research article development of a system to assist...
TRANSCRIPT
Research ArticleDevelopment of a System to Assist AutomaticTranslation of Hand-Drawn Maps into Tactile Graphics andIts Usability Evaluation
Jianjun Chen and Noboru Takagi
Department of Intelligent Systems Design Engineering Toyama Prefectural University 5180 Kurokawa Imizu Toyama 939-0398 Japan
Correspondence should be addressed to Jianjun Chen jianjun0221happy163com
Received 7 April 2014 Accepted 30 June 2014 Published 23 July 2014
Academic Editor Erich Peter Klement
Copyright copy 2014 J Chen and N Takagi This is an open access article distributed under the Creative Commons AttributionLicense which permits unrestricted use distribution and reproduction in any medium provided the original work is properlycited
Tactile graphics are images that use raised surfaces so that a visually impaired person can feel them Tactile maps are used byblind and partially sighted people when navigating around an environment and they are also used prior to a visit for orientationpurposes Since the ability to read tactile graphics deeply depends on individuals providing tactile graphics individually is neededThis implies that producing tactile graphics should be as simple as possible Based on this background we are developing a systemfor automating production of tactile maps from hand-drawn figures In this paper we first present a pattern recognition methodfor hand-drawn maps The usability of our system is then evaluated by comparing it with the two different methods to producetactile graphics
1 Introduction
Producing tactile maps is an important effort to bring blindpeople to more self-supported life There have been manystudies of computer-aided systems in order to assist theproduction of tactile graphics [1ndash6] Tactile map automatedcreation system (TMACS) [2 3] for example has beendeveloped to produce tactile maps automatically It is a webapplication and produces the digital file for a tactile mapfrom the information about two places departure placeand destination TMACS assumes that users can be blindand so it produces the tactile map automatically from themap database if a user only provides a departure place anddestination to the system However tactile maps producedby TMACS are sometimes difficult to read for the blindbecause it is possible to include unnecessary information inthe tactile maps Further Geospatial Information Authorityof Japan has also developed a tactile map production system[4] This system assumes that users are sighted people and itis totally provided as a GUI application Operating this GUIapplication is not easy for users who are not familiar withcomputers
Based on the background above we are now developinga system for automating production of tactile maps In thetactile map production method using our system a sighteduser first draws manually a hand-drawn map using a penciland paper and the map is then converted to a digital imageusing an image scanner or a digital camera Finally by usingour system the digital image is recognized and translated intodigital files which are available to produce the tactile mapsOur system chooses the Edel [7] and scalable vector graphics(SVG) [8] documents as the output file formats Here Edel isa software system to create digital files available to producetactile graphics by Braille embossers and this system is usedwidely in Japanese blind schools The advantages for oursystem are as follows
(1) Sighted users can draw maps by using pencils andpapers no computer operation is needed when draw-ing maps
(2) Sighted users can easily add necessary informationand remove unnecessary information This impliesthat providing tactile maps individually is not timeconsuming
Hindawi Publishing CorporationAdvances in Fuzzy SystemsVolume 2014 Article ID 357380 11 pageshttpdxdoiorg1011552014357380
2 Advances in Fuzzy Systems
(a) Departureplace (b) Destination (c) Trafficsignal (d) Route (e) Railway (f) Landmark (g) Street
Figure 1 Object types and symbols
(a) (b)
Figure 2 Examples for hand-drawn maps
Original color image
Object classification
Segmentation
Preprocessing
Shape classification
Traffic signal detection
SVGEdel documents production
Figure 3 Outline for hand-drawn map translation
It is important to evaluate the usability for the methodusing our systemThe evaluationwas done by comparingwiththe methods using different software systems
The remainder of this paper is constructed as fol-lows Section 2 outlines our method and shows some ofthe basic procedures Object classification is explained inSection 3 fuzzy inference is applied to realize the classifica-tion methods The production of SVG and Edel documentsis described in Section 4 The classification accuracy for ourmethod is shown in Section 5 In Section 6 we discuss theusability evaluation for our system and conclude the study inSection 7
2 Outline for Our Method andBasic Procedures
In order to facilitate hand-drawnmap recognition we assumethe following conditions for drawing maps
(1) A hand-drawn map consists of the following objecttypes departure place destination traffic signalroute railway landmark and street The symbols forobjects are summarized in Figure 1
(2) A hand-drawnmap is a line drawing except for trafficsignals and the bullet for the destination symbol
(3) A landmark object is represented by a polygon andforms a connected component Further it does notconnect to any other objects
Figure 2 shows examples for hand-drawnmaps Inputtingthe digital image for a hand-drawn map to our system itoutputs two digital files for tactile maps the SVG and Edeldocuments The procedure for hand-drawn map translationis outlined in Figure 3
In our tactile map production method a map is firstdrawn manually using a pencil and paper Then the hand-drawn map is captured by an image scanner or a digitalcamera Finally our system translates the digital image for thehand-drawn map into SVG and Edel documents producingthe tactile maps
There are various new technologies applied in hand-drawn sketch recognition For example Broelemann et al [9]developed a method for automatic street graph constructionof hand-drawn maps but this method is restricted to detec-tion of streets The sketch recognition methods introduced
Advances in Fuzzy Systems 3
(a) Original image (b) Thresholded image by Otsursquos method
(c) Thresholded top-hat image
Figure 4 Using the top-hat transformation for shading correction the top-hat transformation corrects the effects of nonuniformillumination and as a result the fine binary image is created using Otsursquos method
in [10ndash14] cannot be used to recognize the hand-drawn mapdue to their limitations So a new method is proposed torecognize the hand-drawn map in this paper
In the remainder of this section preprocessing trafficsignal detection segmentation and shape classification areexplained
21 Preprocessing and Traffic Signal Detection The prepro-cessing consists of the following three procedures
(1) An original color image is converted to a grayscaleimage 119868 by the formula
119868 = 02989 sdot 119877 + 05866 sdot 119866 + 01145 sdot 119861 (1)
where 119877 119866 and 119861 are the intensities for red greenand blue of the original image
(2) The two noise reduction methods are applied to thegrayscale image
(a) The median filter with 3 times 3 window is firstapplied to eliminate salt-and-pepper noise
(b) a grayscale morphological top-hat transforma-tion [15] is applied to correct the effects ofnonuniform illumination Here a disk structur-ing element of radius 3 pixels is used for thetop-hat transformation The effect for a top-hattransformation is explained in Figure 4
(3) The grayscale image is converted to a binary imageusing Otsursquos method [16]
Mathematical morphology is applied to extract trafficsignals from the binary image First an erosion operatoris performed to the binary image for three times and adilation operator is then applied to the eroded image for threetimes The structuring element is a disk of radius 3 pixelsThe erosion operation eliminates thinner line segments butit does not remove traffic signals completely The dilationoperation then detects all the traffic signals By this detectionthe bullet for the destination symbol is also extracted
22 Segmentation and Shape Classification After detectingtraffic signals we segment objects into fragments The fourprocedures are introduced to divide objects (1) thinning (2)short branch removal (3) feature point detection and (4)dividing
(1) Hilditchrsquos thinning algorithm [17] is first applied to thebinary image and the skeleton image is obtained
(2) Short branch removal is as follows Every shortbranch of the skeleton is removed if its length isshorter than a threshold value
(3) Feature point detection is as follows Intersections andcorner points are detected in this stage A 3times3 slidingwindow is applied to the skeleton image in order todetect intersectionsThemethod for detecting cornerpoints is explained below
(4) Dividing is as follows By removing all intersectionsand corner points in the skeleton the skeleton isdivided into fragments It is characteristic that every
4 Advances in Fuzzy Systems
Ox
y
P1(x1 y1)
P2(x2 y2)
P3(x3 y3)
Figure 5 Piecewise linear approximation
fragment has two endpoints or no endpoint Frag-ments are called elements in this paper
Corner Point Detection There have been many studies fordetecting corner points in digital images [18ndash20] We intro-duce a new method to precisely detect corner points forhand-drawn figures This method first finds a piecewiselinear approximation (PL approximation for short) for adigital curve this approximation is detected by Wall andDanielssonrsquos algorithm [21] Let 119874 119875
1 and 119875
2be the origin
and the two different points on the 119909119910 plane (see Figure 5) Ifthe ratio of the area for the triangle119874119875
11198752to the length of the
line segment1198741198752exceeds a threshold value then the point 119875
1
is chosen as a point of the PL approximation if not the nextpoint 119875
2is examined Let us consider a digital curve denoted
by a sequence of the points119862 = 1198750 1198751 119875
119899minus1 TheWall and
Danielssonrsquos algorithm is then presented below
Step 1 Set 119894 larr 1 and choose the point 1198750as the first starting
point for the PL approximation
Step 2 Let 119874 be the point 119875119894minus1
and set 119890119894larr 0
Step 3 Set 119894 larr 119894 + 1 and calculate 119890119894and ℓ119894 Consider
119890119894larr997888 119890119894minus1
+ Δ119890119894
ℓ119894larr997888 radic119909
2
119894+ 1199102
119894
(2)
where Δ119890119894is the value 119909
119894minus1119910119894minus 119909119894119910119894minus1
which is the singed areafor the parallelogram spanned by the two vectors 997888997888997888997888rarr119874119875
119894minus1and
997888997888rarr119874119875119894
Step 4 If |119890119894| le 119905 sdot ℓ
119894holds then go to Step 3 if not choose
119875119894minus1
as a point of the PL approximation and then go to Step 2Here 119905 is a threshold value
The above algorithm cannot detect corner points cor-rectly However if a corner point exists it must be close toa point of the PL approximation So we improve the Walland Danielssonrsquos algorithm by exchanging Step 4 with thefollowing Step 41015840 so that the algorithm is able to preciselydetect corner points of a hand-drawn digital curve
So we introduce two procedures explained below andour corner detection procedure is the one that is obtained byexchanging Step 4 with the following Step 41015840
Step 41015840 If |119890119894| le 119905 sdot ℓ
119894holds then go to Step 3 if not the
point 119875119894minus1
is detected as a point of the PL approximation andperform the following process to test whether a corner pointexists near the point 119875
119894minus1
(1) Conduct the following procedure 1 whose input is119875119894minus1
If this procedure returns a point choose thispoint as a corner point of the digital curve119862 and thengo to Step 2
(2) Conduct the following procedure 2 whose input is119875119894minus1
If this procedure returns a point choose thispoint as a corner point of the digital curve119862 and thengo to Step 2
Procedure 1 aims to detect a sharp corner point Thisprocedure is performed in the following steps
Step 1 Let 119875119894minus119904
(119904 gt 1) be a point of the digital curve 119862satisfying the following two conditions
(1) 119875119894minus119904
has been chosen as a point of the PL approxima-tion
(2) 119875119894minus1
is the next PL approximation point to 119875119894minus119904
Step 2 For every point 119875119894minus119904+119896
of the digital curve 119862 between119875119894minus119904
and 119875119894minus1
(119896 = 1 2 119904 minus 1) calculate the distancedenoted by ℓ
119896 between 119875
119894minus119904and 119875
119894minus119904+119896 Then let 119871 be the
sequence ℓ1 ℓ2 ℓ
119904minus1
Step 3 For the sequence 119871 if the distances for119871 monotonously increase in the middle and themonotonously decrease to the end that is the inequalitiesℓ1lt ℓ2lt sdot sdot sdot lt ℓ
119902gt ℓ119902+1
gt sdot sdot sdot gt ℓ119904minus1
hold then the point119875119894minus119904+119902
is determined as a corner point of the digital curve 119862Next the procedure 2 is explained it is motivated by the
method for the literature [22] Remember that the point 119875119894minus1
has been chosen as a point of the PL approximation
Step 1 For every 119875119894minus119896
(119896 = 1 2 21) calculate thesharpness 119904(119875
119894minus119896) the procedure to calculate sharpness is
explained below
Step 2 For the sequence of the sharpnesses119904(119875119894minus1) 119904(119875119894minus2) 119904(119875
119894minus21) calculate the simple moving
averages smv(119875119894minus1) smv(119875
119894minus21)
Step 3 If the maximum value of smv(119875119894minus1) smv(119875
119894minus21)
exceeds a threshold value then the point 119875119894minus119896
which gives themaximum value is determined as a corner point of the digitalcurve 119862 if not we decide that no corner point exists near thepoint 119875
119894minus1
The calculation for the sharpness 119904(119875) for a point 119875 is asfollows We first detect two subsequences of points denotedby 119865 = 119875
1198961
119875119896119899
and 119861 = 1198751199051
119875119905119898
in the following way(see Figure 6) We visit the points forward from 119875 and select
Advances in Fuzzy Systems 5
Direction
Digital curve
B F
Pt119898+1
Pt119898
Pt1Pt2Pt3
P
Pk119899+1
Pk119899
Pk1Pk2Pk3
Figure 6 Sharpness
a point 119875119896119895
the distance between 119875119896119895
and 119875 is more than orequal to 2 pixels and less than or equal to 10 pixels 119875
1198961
is thefirst point satisfying this condition We select points until thecondition breaks119875
119896119899
is the last point satisfying the conditionThe subsequence 119861 is obtained similarly while visiting thepoints backward from 119875 The sharpness 119904(119875) is then definedas the following formula
119904 (119875) =1
119899119898
119899
sum
119903=1
119898
sum
119904=1
(the angle of ang119875119896119903
119875119875119905119904
) (3)
Next shape classification is explained Every element isclassified into one of the four shapes straight line circlecircular arc and curve This classification is done by themethod of least squares That is we minimize the sum ofthe squared Euclidean distances between the points of theelement and the corresponding points on the model Theelement is then classified as the shape for the model if thesum is smaller than a threshold value A curve is expressed bya piecewise cubic Bezier curve we will describe cubic Beziercurves in Section 4
3 Object Classification
In this section object classification is described First ofall a landmark object is classified by the following simpleprocedure if an object is isolated and closed and has morethan 3 corners the object is classified as a landmark Exceptfor landmark classification fuzzy inference is applied todesign the classification methods The fuzzy inference sys-tems introduced in this paper are constructed by Mamdanirsquosfuzzy inference method [23] but the minimum operator isexchanged with the product operator To detect route andrailway objects it is needed to find arrows and crosses Soelements are first grouped into a single cluster if they areconnected to the same intersection Every cluster is thenexamined if it is an arrow or a cross Note that an elementcan belong to different two clusters when the two endpointsof the element connect to the different intersections
31 Arrow Classification An arrow consists of four or threestraight line elements (see Figures 7(a) and 7(b)) We applytwo fuzzy inference systems to calculate the similarity forarrow The following description is the outline of the firstfuzzy inference system
e1
e2
e3
e4
(a)
e1
e2
e3
(b)
12059311205932
1205933 1205934
e1
e2
e3
e4
(c)
1205931
1205932
1205933
e1
e2
e3
(d)
Figure 7 Elements for arrows
(1) We first choose a cluster 119864 which includes fourstraight line elements denoted by 119890
1 1198902 1198903 and 119890
4
The four angles 1205931 1205932 1205933 and 120593
4are then measured
(see Figure 7(c))
(2) Next the following three attribute values are calcu-lated
(1) the standard deviation of 1205931 1205932 1205933 and 120593
4 it is
denoted by 1199091
(2) the average value of |1205931minus 1205934| and |120593
2minus 1205933| it is
denoted by 1199092
(3) the difference between the two lengths of theelements 119890
2and 1198904 it is denoted by 119909
3
(3) The fuzzy inference system is applied to 1199091 1199092 and
1199093 The fuzzy if-then rules are denoted below and
the membership functions are shown in Figure 8 Wethen have a value 119904 of [0 1] from the system 119904 impliesthe similarity for 119864
Rule 1 If 1199091is large 119909
2is small and 119909
3is small then it is
plausible that 119864 is an arrow
Rule 2 If 1199091is small then it is implausible that 119864 is an arrow
Rule 3 If 1199092is large then it is implausible that 119864 is an arrow
Rule 4 If 1199093is large then it is implausible that 119864 is an arrow
The second fuzzy inference system is applied to a cluster119864 which includes three straight line elements This systemhas the following three input attributes
(1) the standard deviation for 1205931 1205932 and 120593
3(see
Figure 7(d)) it is denoted by 1199091
(2) the difference between 1205932and 120593
3 it is denoted by 119909
2
(3) the difference between the two lengths of the elements1198901and 1198903 it is denoted by 119909
3
6 Advances in Fuzzy Systems
y = 1205831large (x1)y = 1205831small (x1)
y
10
0 10 30x1
(a) Membership functions for 1199091
y = 1205832large (x2)y = 1205832small (x2)
y
10
0 10 15x2
(b) Membership functions for 1199092
y = 1205833large (x3)y = 1205833small (x3)
y
10
0 10 20x3
(c) Membership functions for 1199093
y = 120583plau (x)y = 120583implau (x)
y
10
02 05 08x
0
(d) Membership functions for consequence
Figure 8Membership functions 120583119894small and 120583119894
large (119894 = 1 2 3) aremembership functions for the fuzzy sets small and large in Rule 119895 (119895 = 1 2 3)120583plau and 120583implau are membership functions for the fuzzy sets plausible and implausible in the consequence
e1
e2
e3
e4
(a)
e1
e2
e3
e4
12059311205932
1205933 1205934
(b)
Figure 9 Elements for cross
The following rules are the fuzzy if-then rules The member-ship functions are omitted
Rule 1 If 1199091is large 119909
2is small and 119909
3is small then it is
plausible that 119864 is an arrow
Rule 2 If 1199091is small then it is implausible that 119864 is an arrow
Rule 3 If 1199092is large then it is implausible that 119864 is an arrow
Rule 4 If 1199093is large then it is implausible that 119864 is an arrow
32 Cross Classification A cross consists of four straight lineelements (see Figure 9(a)) A cluster 119864 is applied to the fuzzyinference system for cross classification if 119864 includes four
straight line elements 1198901 1198902 1198903 and 119890
4 This fuzzy inference
system requires the following two input attributes(1) the standard deviation of 120593
1 1205932 1205933 and 120593
4(see
Figure 9(b)) it is denoted by 1199091
(2) the standard deviation for 1205931+ 1205932 1205932+ 1205933 1205933+ 1205934
and 1205931+ 1205934 it is denoted by 119909
2
The fuzzy if-then rules are given belowwhile themembershipfunctions are omitted
Rule 1 If 1199091is small and 119909
2is small then it is plausible that 119864
is a cross
Rule 2 If 1199091is large then it is implausible that 119864 is a cross
Rule 3 If 1199092is large then it is implausible that 119864 is a cross
If the similarity for a cluster obtained by the crossclassification is larger than a threshold value the cluster isclassified as a cross Similarly if the similarity from the arrowclassification exceeds a threshold value the cluster is classifiedas an arrow
33 Route and Railway Classification Route and railwayclassifications are conducted by the following way
(1) We first detect a sequence denoted by 119876 of clusterssuch that every two adjacent clusters include a com-mon straight line element (see Figure 10)
Advances in Fuzzy Systems 7
The principal line Arrowheads
e1 e2
(a) Route with three arrows the elements 1198901 and 1198902are common to the left and the right arrows
The principal line
e1 e2
Auxiliary lines
(b) Railway with three crosses the elements 1198901 and 1198902are common to the left and the right crosses
Figure 10 Route and railway symbols
Pi+1(xi+1 yi+1)
Pi(xi yi)
eiminus1
ei
120593iminus1
120593i
Piminus1(ximinus1 yiminus1)
Figure 11 Curvature for digital curve
(2) If the sequence 119876 consists of arrows then the routeclassification is executed to calculate the similaritydenoted by 119904
1 for the sequence119876 If 119904
1is larger than a
threshold value the sequence119876 is classified as a route(3) If the sequence119876 consists of crosses then the railway
classification is executed and we obtain the similaritydenoted by 119904
2 for the sequence 119876 If 119904
2exceeds a
threshold value the sequence 119876 is classified as arailway
Fuzzy inference systems are applied to compute thetwo similarities To calculate the attribute values for thefuzzy inference systems we detect the principal line andthe arrowhead (or the auxiliary lines) for the sequence 119876 asfollows If two straight line elements 119890
1and 119890
2 satisfy the
following two geometric characteristics it is plausible that 1198901
and 1198902are part of the same straight line
(1) The elements 1198901and 1198902are connected at an intersec-
tion 119875(2) The curvature at the intersection is small
Let119862 be a digital curve and let119875119894be a point on119862 A curvature
120581 of 119862 at 119875119894is then defined as the subtraction |120593
119894minus 120593119894minus1| (see
Figure 11) that is
120581 =1003816100381610038161003816120593119894 minus 120593119894minus1
1003816100381610038161003816 (4)
where 120593119894is the measure of an angle which is formed between
the 119909-axis and the line segment from 119875119894to 119875119894+1
and 120593119894minus1
isalso the measure of an angle which is formed between the 119909-axis and the line segment from 119875
119894minus1to 119875119894[24] Two adjacent
elements 119890119894minus1
and 119890119894are merged into a single straight line
segment if the curvature between 119890119894minus1
and 119890119894is less than a
threshold value After that we extract the principal line andthe arrowheads (or the auxiliary lines) for the sequence 119876
After extracting the principle line and the arrowhead (orthe auxiliary lines) for the sequence 119876 two fuzzy inferencesystems are applied to classify the sequence 119876 as a routeor railway The first fuzzy inference system is for routeclassification It has the following two input attributes
(1) the standard deviation for the lengths of elements inthe principal line it is denoted by 119909
1
(2) the standard deviation for lengths of elements in thearrowheads it is denoted by 119909
2
The fuzzy if-then rules for the first fuzzy inferencesystem are denoted below but the membership functions areomitted
Rule 1 If 1199091is small and 119909
2is small then it is plausible that119876
is a route
Rule 2 If 1199091is large then it is implausible that 119876 is a route
Rule 3 If 1199092is large then it is implausible that 119876 is a route
The second fuzzy inference system is for railway classifi-cation and the description for the second system is omittedbecause it is similar to the first system
4 SVG and Edel Documents Production
In our method a figure consists of elements whose shapesare as follows straight line circle circular arc and curveAs described in Section 22 the shape for an element isdetermined by the method of least squares If an element hasbeen classified as a curve then this element is expressed bya piecewise cubic Bezier curve the piecewise cubic Beziercurve is detected by an algorithmbased on themethodof leastsquares [25]
A Bezier curve 119876(119905) of degree 119899 is defined as
119876 (119905) =
119899
sum
119894=0
119881119894119861119899
119894(119905) (119905 isin [0 1]) (5)
where the 119881119894s are the control points and the 119861119899
119894s are the
Bernstein polynomials
119861119899
119894(119905) = (
119899
119894) 119905119894(1 minus 119905)
119899minus119894(119894 = 0 1 119899) (6)
8 Advances in Fuzzy Systems
Figure 12 A single cubic Bezier segment
0
2
4
6
8
0h lt 2h 2ndash5h 5ndash10h 10hlt
Figure 13 Experience in computers the vertical axis indicates thenumber of participants
Equation (5) is called a cubic Bezier curve when 119899 = 3Figure 12 shows an example of cubic Bezier curves the fourpoints119881
011988111198812 and119881
3are the control points the thick curve
is the cubic Bezier curve and the polygon expresses the oneconstructed by the four control points
SVG has the formats to express the four shapes andtherefore we can create an SVG document for a hand-drawnmap directly once we have detected the shapes for all theelements of the map On the other hand an Edel documentis a collection of embossed dots So it is easy to create anEdel document once we have detected the shapes for all theelements of a hand-drawn map
5 Experimental Results
This section describes the accuracy for our classificationsystem Five participants drew 15 maps using pencils andpapers These maps were then captured by using an imagescanner the resolution of the scannerwas set to 100 dpiThesedigital images were saved as 24-bits bitmap images The sizesof the images are in 1 169times850 pixelsWe havemeasured theaccuracy for our classification system using precision recalland f -measure
In the 15 maps there are 50 traffic signals 15 destinations15 departure places and 40 landmarks The classificationaccuracy for our system is summarized in Table 1 We canconclude that our system can produce the Edel and SVGdocuments almost correctly from hand-drawn maps
0
2
4
6
8
Freq
uent
ly
Som
etim
es
Occ
asio
nally
Seld
om
Nev
er
Figure 14 Experience in PPT the vertical axis indicates the numberof participants
Table 1 Experimental results are as follows 119873 is the number of allobjects for the symbol119879 is the number of objects correctly classifiedas the symbol 119862 is the number of objects classified as the symbol 119875is precision 119877 is recall and 119865 is 119865-measure
Symbol 119873 119879 119862 119875 () 119877 () 119865 ()Traffic signal 50 50 52 962 100 981Destination 15 15 15 100 100 100Departure place 15 14 14 100 933 965Landmark 40 38 38 100 950 974Cross 130 125 125 100 962 981Arrow 122 121 121 100 992 996Route 15 14 16 875 933 903Railway 28 28 28 100 100 100
6 Usability Evaluation for Our System
61 Method We have studied usability evaluation for oursystemThirteen participants 12 males and 1 female aged 20participated in this investigation all of them are the third-year university students To evaluate the usability for oursystem we selected two common methods for productionof tactile graphics one is the method to use the softwaresystem Edel which assists us to draw diagram images fortactile graphics produced by Braille embossers and anotherone is the method using swell papers In the second methoda diagram image is transferred to a swell paper by a printerand so forth A swell paper has been coated with thermallyfoamed microcapsules that respond to irradiation and causethe dark image on the paper to swell creating the tactilegraphic Microsoft PowerPoint (PPT) was selected to draw adiagram image on a swell paper because this software systemis commonly used in the universities in Japan
Before starting the experiment we asked the followingquestions Q1 Q2 and Q3 to all the participants and theresults for the questionsQ1 andQ3 are summarized in Figures13 and 14 We omit to show the result for the question Q2because all the participants have no experience in using EdelAll the participants are familiar with using computers andhave much experience in using PTT
Advances in Fuzzy Systems 9
(a) (b)
(c) (d)
(e) (f)
Figure 15 Maps for Sessions 1 and 2 (a) Edel map for Session 1 (b) PPT map for Session 1 (c) hand-drawn map for Session 1 (d) Edel mapfor Session 2 (e) PPT map for Session 2 and (f) hand-drawn map for Session 2
Q1 How long do you use computers every weekQ2 Have you ever created figures by using EdelQ3 Have you ever created figures by using PPT
We had two sessions in this investigation For the firstsession a participant created the map diagram shown inFigures 15(a)ndash15(c) the digital files were produced by usingEdel PPT and our system Note that in our system a userfirst draws a map manually using a pencil and paper thenconverts the hand-drawn map to a bitmap image using animage scanner and finally the bitmap image is transformed
into the two digital files the Edel and SVG documents Afterconducting all tasks the participant was asked to answer thefollowing two questions Q4 and Q5
Q4 Were you able to easily create the digital file(s) for thetactile map
Q5 Do you want to use this software system to create thedigital file for a tactile map
After the first session the participant had 5 minutes fortraining the software systems by himselfherself The secondsession followed this learning session In the second session
10 Advances in Fuzzy Systems
the participant conducted similar tasks to the first sessionbut the maps were those shown in Figures 15(d)ndash15(f) Aftercreating all digital files the participant was asked to answertwo questions Q4 and Q5 again and was also asked to answerthe following question Q6
Q6 Were you able to draw the map as you like
The orders of the software systems were determined ran-domly in both the first and the second sessions
62 Results We applied two-way ANOVA to the answersfor the 2 questions Q4 and Q5 The first factor denoted byFactor 1 is the difference in systems and the second factordenoted by Factor 2 is the difference in learning Factor 1includes three levels Edel PPT and our system and Factor2 includes two levels before learning and after learning Theresults from the two-way ANOVA for Q4 shows that therewas no significant interactions between Factor 1 and Factor2 (119865(2 72) = 092 119875 = 040) Furthermore there was asignificant difference for Factor 1 (119865(2 72) = 1926119875 lt 001)while no significant difference existed for Factor 2 (119865(1 72) =158 119875 = 021)The results for Q5 are similar to those for Q4That is there was no significant interaction between Factor1 and Factor 2 (119865(2 72) = 064 119875 = 053) there was asignificant difference for Factor 1 (119865(2 72) = 1400119875 lt 001)and therewas no significant difference for Factor 2 (119865(1 72) =132 119875 = 025)
We then conducted a multiple comparison test for theanswers of the questions Q4 and Q5 we selected Tukeyrsquoshonestly significant difference (HSD) criterion as themultiplecomparison test For Q4 there was a significant differencebetween Edel and our system (119875 lt 001) and there was alsoa significant difference between PPT and our system (119875 lt
001) however there was no significant difference betweenEdel and PPT (see Figure 16) Furthermore for Q5 therewas a significant difference between Edel and our system(119875 lt 001) and therewas also a significant difference betweenPPT and our system (119875 lt 005) however we observed amarginally significant difference between Edel and PPT (119875 lt01) (see Figure 17)
Lastly we applied a one-way ANOVA to the answersfor the question Q6 the factor of this one-way ANOVA isthe difference between systems The results are shown inFigure 18 and the results for the one-way ANOVA show thatthere was a marginally significant difference (119865(2 36) = 259119875 = 0089) We then applied Tukeyrsquos HSD and we observeda marginally significant difference between Edel and oursystem
As a result of the discussion above we can concludethat users feel that our method is easier to produce tactilemaps than the conventional twomethods Furthermore mapimages produced by our system are visually as fine as mapimages produced by Edel and PPT
7 Conclusions
In this paper a computer-aided system for automating pro-duction of tactile graphics from hand-drawn maps has been
0
2
4
6
Edel PPT Our System
Scor
e
Before LearningAfter Learning
lowastlowastlowastlowast
Figure 16 Results for Q4 the average and standard error for eachsystem (lowastlowast119875 lt 001)
0
1
2
3
4
Edel PPT Our System
Scor
e
lowastlowastlowast
Before LearningAfter Learning
Figure 17 Results for Q5 the average and standard error for eachsystem (lowastlowast119875 lt 001 lowast119875 lt 005)
developed The system includes the following four mainprocedures (1) preprocessing (2) segmentation (3) patternrecognition based on fuzzy inference and (4) SVG and Edeldocuments production The preprocessing is applied to aninitial hand-drawn map in order to remove noise and thena binary image is obtained In the segmentation procedurethe binary image is first skeletonized and then the skeletonis segmented into elements by eliminating intersections andcorner points Then a pattern recognition method is appliedto recognize symbols Lastly SVG and Edel documents arecreated to save the recognition results The usability for oursystem was evaluated by comparing our method with the twoconventional systems for creating tactile maps The resultsshow that our system is easier to create tactile maps than theother two systems
In our fuzzy inference systems triangular and trapezoidalmembership functions are applied to define the fuzzy setsdue to their simplicity The shapes of membership functionsinfluence the accuracy of recognition results How to choosean optimal membership function is our future work In thisstudy we do not assume a map includes character stringsHowever adding character strings to a map is very important
Advances in Fuzzy Systems 11
0
1
2
3
4
Edel PPT Our System
Scor
e
Figure 18 Results for Q6 the average and standard error for eachsystem
because it increases the comprehension for the map In thepresent work the types of symbols used in hand-drawnmaps are restricted Developing a system for automatingtranslation of hand-drawn maps with character string andvarious symbols is one of our future works
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This work was supported by the JSPS Grant-in-Aid forScientific Research (C) no 24501198 The authors thank DrKeisuke Ido for giving much advice to the experiments ofusability evaluation
References
[1] J A Miele S Landau and D Gilden ldquoTalking TMAPautomated generation of audio-tactile maps using Smith-Kettlewellrsquos TMAP softwarerdquo The British Journal of VisualImpairment vol 24 no 2 pp 93ndash100 2006
[2] T Watanabe T Yamaguchi K Watanabe et al ldquoDevelopmentand evaluation of a tactile map automated creation systemaccessible to blind personsrdquo IEICE Transactions on Informationand Systems vol J94-D no 10 pp 1652ndash1663 2011 (Japanese)
[3] TMACS httptmacsengniigata-uacjptmacs-dev[4] httptenpuchizugsigojpshokuchizu[5] S E Krufka and K E Barner ldquoAutomatic production of tactile
graphics from scalable vector graphicsrdquo in Proceedings of the 7thInternational ACM SIGACCESS Conference on Computers andAccessibility (ASSETS rsquo05) pp 166ndash172 October 2005
[6] R E Ladner M Y Ivory R Rao et al ldquoAutomating tactilegraphics translationrdquo in Proceedings of the 7th InternationalACMSIGACCESSConference onComputer andAccessibility pp150ndash157 New York NY USA October 2005
[7] Edel httpwww7abiglobenejp EDEL-plus[8] J D Eisenberg SVG Essentials OrsquoREILLY 2002[9] K Broelemann X Jiang and A Schwering ldquoAutomatic street
graph construction in sketchmapsrdquo inGraph-Based Representa-tions in Pattern Recognition X Jiang M Ferrer and A Torsello
Eds vol 6658 of Lecture Notes in Computer Science pp 275ndash284 Springer Berlin Germany 2011
[10] L B Kara and T F Stahovich ldquoAn image-based trainablesymbol recognizer for hand-drawn sketchesrdquo Computers ampGraphics vol 29 no 4 pp 501ndash517 2005
[11] B Edwards and V Chandran ldquoMachine recognition of hand-drawn circuit diagramsrdquo in Proceedings of the IEEE InterntionalConference onAcoustics Speech and Signal Processing vol 6 pp3618ndash3621 June 2000
[12] P Sala ldquoA recognition system for symbols of electronic com-ponents in hand-written circuit diagramsrdquo CSC 2515-MachineLearning Project Report 2004
[13] Y Qiao and M Yasuhara ldquoRecovering dynamic informationfrom static handwritten imagesrdquo in Proceedings of the 9th Inter-national Workshop on Frontiers in Handwriting Recognition pp118ndash123 jpn October 2004
[14] M Notowidigdo and R C Miller ldquoOff-line sketch interpreta-tionrdquo in Proceedings of the AAAI Fall Symposium on MakingPen-Based Interaction Intelligent and Natural pp 120ndash126Arlington Va USA April 2004
[15] R J Gonzalez and R E Woods Digital Image ProcessingPearson Education Upper Saddle River NJ USA 3rd edition2007
[16] N Otsu ldquoA threshold selection method from gray-level his-togramsrdquo IEEE Transactions on Systems Man and Cyberneticsvol 9 no 1 pp 62ndash66 1979
[17] C J Hilditch ldquoLinear skeletons from square cupboardsrdquo inMachine Intelligence Vol IV B Meltzer andDMichie Eds pp403ndash420 Elsevier New York NY USA 1969
[18] S Hermann and R Klette ldquoGlobal curvature estimation for cor-ner detectionrdquo Communication and Information TechnologyResearch Technical Report 171 2005
[19] N Nain V Laxmi B Bhadviya and A Gopal ldquoCorner detec-tion using difference chain code as curvaturerdquo in Proceedingsof the 3rd IEEE International Conference on Signal ImageTechnologies and Internet Based Systems (SITIS 07) pp 821ndash825Shanghai China December 2007
[20] B Kerautret J Lachaud and B Naegel ldquoCurvature basedcorner detector for discrete noisy and multi-scale contoursrdquoInternational Journal of Shape Modeling vol 14 no 2 pp 127ndash145 2008
[21] K Wall and P Danielsson ldquoA fast sequential method forpolygonal approximation of digitized curvesrdquoComputer VisionGraphics amp Image Processing vol 28 no 2 pp 220ndash227 1984
[22] D Chetverikov ldquoA simple and efficient algorithm for detectionof high curvature points in planar curvesrdquo inComputer Analysisof Images and Patterns vol 2756 of Lecture Notes in ComputerScience pp 746ndash753 Springer Berlin Germany 2003
[23] T Terano K Asai and M Sugeno Fuzzy Systems and ItsApplications Academic Press Boston Mass USA 2nd edition1992
[24] N Ono and R Takiyama ldquoOn calculations of curvature ofsampled curvesrdquo Technical Report of IEICE IE93-74 1993(Japanese)
[25] P J Schneider ldquoAn algorithm for automatically fitting digitizedcurvesrdquo in Graphics Gems A S Glassner Ed pp 612ndash625Academic Press 1990
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
2 Advances in Fuzzy Systems
(a) Departureplace (b) Destination (c) Trafficsignal (d) Route (e) Railway (f) Landmark (g) Street
Figure 1 Object types and symbols
(a) (b)
Figure 2 Examples for hand-drawn maps
Original color image
Object classification
Segmentation
Preprocessing
Shape classification
Traffic signal detection
SVGEdel documents production
Figure 3 Outline for hand-drawn map translation
It is important to evaluate the usability for the methodusing our systemThe evaluationwas done by comparingwiththe methods using different software systems
The remainder of this paper is constructed as fol-lows Section 2 outlines our method and shows some ofthe basic procedures Object classification is explained inSection 3 fuzzy inference is applied to realize the classifica-tion methods The production of SVG and Edel documentsis described in Section 4 The classification accuracy for ourmethod is shown in Section 5 In Section 6 we discuss theusability evaluation for our system and conclude the study inSection 7
2 Outline for Our Method andBasic Procedures
In order to facilitate hand-drawnmap recognition we assumethe following conditions for drawing maps
(1) A hand-drawn map consists of the following objecttypes departure place destination traffic signalroute railway landmark and street The symbols forobjects are summarized in Figure 1
(2) A hand-drawnmap is a line drawing except for trafficsignals and the bullet for the destination symbol
(3) A landmark object is represented by a polygon andforms a connected component Further it does notconnect to any other objects
Figure 2 shows examples for hand-drawnmaps Inputtingthe digital image for a hand-drawn map to our system itoutputs two digital files for tactile maps the SVG and Edeldocuments The procedure for hand-drawn map translationis outlined in Figure 3
In our tactile map production method a map is firstdrawn manually using a pencil and paper Then the hand-drawn map is captured by an image scanner or a digitalcamera Finally our system translates the digital image for thehand-drawn map into SVG and Edel documents producingthe tactile maps
There are various new technologies applied in hand-drawn sketch recognition For example Broelemann et al [9]developed a method for automatic street graph constructionof hand-drawn maps but this method is restricted to detec-tion of streets The sketch recognition methods introduced
Advances in Fuzzy Systems 3
(a) Original image (b) Thresholded image by Otsursquos method
(c) Thresholded top-hat image
Figure 4 Using the top-hat transformation for shading correction the top-hat transformation corrects the effects of nonuniformillumination and as a result the fine binary image is created using Otsursquos method
in [10ndash14] cannot be used to recognize the hand-drawn mapdue to their limitations So a new method is proposed torecognize the hand-drawn map in this paper
In the remainder of this section preprocessing trafficsignal detection segmentation and shape classification areexplained
21 Preprocessing and Traffic Signal Detection The prepro-cessing consists of the following three procedures
(1) An original color image is converted to a grayscaleimage 119868 by the formula
119868 = 02989 sdot 119877 + 05866 sdot 119866 + 01145 sdot 119861 (1)
where 119877 119866 and 119861 are the intensities for red greenand blue of the original image
(2) The two noise reduction methods are applied to thegrayscale image
(a) The median filter with 3 times 3 window is firstapplied to eliminate salt-and-pepper noise
(b) a grayscale morphological top-hat transforma-tion [15] is applied to correct the effects ofnonuniform illumination Here a disk structur-ing element of radius 3 pixels is used for thetop-hat transformation The effect for a top-hattransformation is explained in Figure 4
(3) The grayscale image is converted to a binary imageusing Otsursquos method [16]
Mathematical morphology is applied to extract trafficsignals from the binary image First an erosion operatoris performed to the binary image for three times and adilation operator is then applied to the eroded image for threetimes The structuring element is a disk of radius 3 pixelsThe erosion operation eliminates thinner line segments butit does not remove traffic signals completely The dilationoperation then detects all the traffic signals By this detectionthe bullet for the destination symbol is also extracted
22 Segmentation and Shape Classification After detectingtraffic signals we segment objects into fragments The fourprocedures are introduced to divide objects (1) thinning (2)short branch removal (3) feature point detection and (4)dividing
(1) Hilditchrsquos thinning algorithm [17] is first applied to thebinary image and the skeleton image is obtained
(2) Short branch removal is as follows Every shortbranch of the skeleton is removed if its length isshorter than a threshold value
(3) Feature point detection is as follows Intersections andcorner points are detected in this stage A 3times3 slidingwindow is applied to the skeleton image in order todetect intersectionsThemethod for detecting cornerpoints is explained below
(4) Dividing is as follows By removing all intersectionsand corner points in the skeleton the skeleton isdivided into fragments It is characteristic that every
4 Advances in Fuzzy Systems
Ox
y
P1(x1 y1)
P2(x2 y2)
P3(x3 y3)
Figure 5 Piecewise linear approximation
fragment has two endpoints or no endpoint Frag-ments are called elements in this paper
Corner Point Detection There have been many studies fordetecting corner points in digital images [18ndash20] We intro-duce a new method to precisely detect corner points forhand-drawn figures This method first finds a piecewiselinear approximation (PL approximation for short) for adigital curve this approximation is detected by Wall andDanielssonrsquos algorithm [21] Let 119874 119875
1 and 119875
2be the origin
and the two different points on the 119909119910 plane (see Figure 5) Ifthe ratio of the area for the triangle119874119875
11198752to the length of the
line segment1198741198752exceeds a threshold value then the point 119875
1
is chosen as a point of the PL approximation if not the nextpoint 119875
2is examined Let us consider a digital curve denoted
by a sequence of the points119862 = 1198750 1198751 119875
119899minus1 TheWall and
Danielssonrsquos algorithm is then presented below
Step 1 Set 119894 larr 1 and choose the point 1198750as the first starting
point for the PL approximation
Step 2 Let 119874 be the point 119875119894minus1
and set 119890119894larr 0
Step 3 Set 119894 larr 119894 + 1 and calculate 119890119894and ℓ119894 Consider
119890119894larr997888 119890119894minus1
+ Δ119890119894
ℓ119894larr997888 radic119909
2
119894+ 1199102
119894
(2)
where Δ119890119894is the value 119909
119894minus1119910119894minus 119909119894119910119894minus1
which is the singed areafor the parallelogram spanned by the two vectors 997888997888997888997888rarr119874119875
119894minus1and
997888997888rarr119874119875119894
Step 4 If |119890119894| le 119905 sdot ℓ
119894holds then go to Step 3 if not choose
119875119894minus1
as a point of the PL approximation and then go to Step 2Here 119905 is a threshold value
The above algorithm cannot detect corner points cor-rectly However if a corner point exists it must be close toa point of the PL approximation So we improve the Walland Danielssonrsquos algorithm by exchanging Step 4 with thefollowing Step 41015840 so that the algorithm is able to preciselydetect corner points of a hand-drawn digital curve
So we introduce two procedures explained below andour corner detection procedure is the one that is obtained byexchanging Step 4 with the following Step 41015840
Step 41015840 If |119890119894| le 119905 sdot ℓ
119894holds then go to Step 3 if not the
point 119875119894minus1
is detected as a point of the PL approximation andperform the following process to test whether a corner pointexists near the point 119875
119894minus1
(1) Conduct the following procedure 1 whose input is119875119894minus1
If this procedure returns a point choose thispoint as a corner point of the digital curve119862 and thengo to Step 2
(2) Conduct the following procedure 2 whose input is119875119894minus1
If this procedure returns a point choose thispoint as a corner point of the digital curve119862 and thengo to Step 2
Procedure 1 aims to detect a sharp corner point Thisprocedure is performed in the following steps
Step 1 Let 119875119894minus119904
(119904 gt 1) be a point of the digital curve 119862satisfying the following two conditions
(1) 119875119894minus119904
has been chosen as a point of the PL approxima-tion
(2) 119875119894minus1
is the next PL approximation point to 119875119894minus119904
Step 2 For every point 119875119894minus119904+119896
of the digital curve 119862 between119875119894minus119904
and 119875119894minus1
(119896 = 1 2 119904 minus 1) calculate the distancedenoted by ℓ
119896 between 119875
119894minus119904and 119875
119894minus119904+119896 Then let 119871 be the
sequence ℓ1 ℓ2 ℓ
119904minus1
Step 3 For the sequence 119871 if the distances for119871 monotonously increase in the middle and themonotonously decrease to the end that is the inequalitiesℓ1lt ℓ2lt sdot sdot sdot lt ℓ
119902gt ℓ119902+1
gt sdot sdot sdot gt ℓ119904minus1
hold then the point119875119894minus119904+119902
is determined as a corner point of the digital curve 119862Next the procedure 2 is explained it is motivated by the
method for the literature [22] Remember that the point 119875119894minus1
has been chosen as a point of the PL approximation
Step 1 For every 119875119894minus119896
(119896 = 1 2 21) calculate thesharpness 119904(119875
119894minus119896) the procedure to calculate sharpness is
explained below
Step 2 For the sequence of the sharpnesses119904(119875119894minus1) 119904(119875119894minus2) 119904(119875
119894minus21) calculate the simple moving
averages smv(119875119894minus1) smv(119875
119894minus21)
Step 3 If the maximum value of smv(119875119894minus1) smv(119875
119894minus21)
exceeds a threshold value then the point 119875119894minus119896
which gives themaximum value is determined as a corner point of the digitalcurve 119862 if not we decide that no corner point exists near thepoint 119875
119894minus1
The calculation for the sharpness 119904(119875) for a point 119875 is asfollows We first detect two subsequences of points denotedby 119865 = 119875
1198961
119875119896119899
and 119861 = 1198751199051
119875119905119898
in the following way(see Figure 6) We visit the points forward from 119875 and select
Advances in Fuzzy Systems 5
Direction
Digital curve
B F
Pt119898+1
Pt119898
Pt1Pt2Pt3
P
Pk119899+1
Pk119899
Pk1Pk2Pk3
Figure 6 Sharpness
a point 119875119896119895
the distance between 119875119896119895
and 119875 is more than orequal to 2 pixels and less than or equal to 10 pixels 119875
1198961
is thefirst point satisfying this condition We select points until thecondition breaks119875
119896119899
is the last point satisfying the conditionThe subsequence 119861 is obtained similarly while visiting thepoints backward from 119875 The sharpness 119904(119875) is then definedas the following formula
119904 (119875) =1
119899119898
119899
sum
119903=1
119898
sum
119904=1
(the angle of ang119875119896119903
119875119875119905119904
) (3)
Next shape classification is explained Every element isclassified into one of the four shapes straight line circlecircular arc and curve This classification is done by themethod of least squares That is we minimize the sum ofthe squared Euclidean distances between the points of theelement and the corresponding points on the model Theelement is then classified as the shape for the model if thesum is smaller than a threshold value A curve is expressed bya piecewise cubic Bezier curve we will describe cubic Beziercurves in Section 4
3 Object Classification
In this section object classification is described First ofall a landmark object is classified by the following simpleprocedure if an object is isolated and closed and has morethan 3 corners the object is classified as a landmark Exceptfor landmark classification fuzzy inference is applied todesign the classification methods The fuzzy inference sys-tems introduced in this paper are constructed by Mamdanirsquosfuzzy inference method [23] but the minimum operator isexchanged with the product operator To detect route andrailway objects it is needed to find arrows and crosses Soelements are first grouped into a single cluster if they areconnected to the same intersection Every cluster is thenexamined if it is an arrow or a cross Note that an elementcan belong to different two clusters when the two endpointsof the element connect to the different intersections
31 Arrow Classification An arrow consists of four or threestraight line elements (see Figures 7(a) and 7(b)) We applytwo fuzzy inference systems to calculate the similarity forarrow The following description is the outline of the firstfuzzy inference system
e1
e2
e3
e4
(a)
e1
e2
e3
(b)
12059311205932
1205933 1205934
e1
e2
e3
e4
(c)
1205931
1205932
1205933
e1
e2
e3
(d)
Figure 7 Elements for arrows
(1) We first choose a cluster 119864 which includes fourstraight line elements denoted by 119890
1 1198902 1198903 and 119890
4
The four angles 1205931 1205932 1205933 and 120593
4are then measured
(see Figure 7(c))
(2) Next the following three attribute values are calcu-lated
(1) the standard deviation of 1205931 1205932 1205933 and 120593
4 it is
denoted by 1199091
(2) the average value of |1205931minus 1205934| and |120593
2minus 1205933| it is
denoted by 1199092
(3) the difference between the two lengths of theelements 119890
2and 1198904 it is denoted by 119909
3
(3) The fuzzy inference system is applied to 1199091 1199092 and
1199093 The fuzzy if-then rules are denoted below and
the membership functions are shown in Figure 8 Wethen have a value 119904 of [0 1] from the system 119904 impliesthe similarity for 119864
Rule 1 If 1199091is large 119909
2is small and 119909
3is small then it is
plausible that 119864 is an arrow
Rule 2 If 1199091is small then it is implausible that 119864 is an arrow
Rule 3 If 1199092is large then it is implausible that 119864 is an arrow
Rule 4 If 1199093is large then it is implausible that 119864 is an arrow
The second fuzzy inference system is applied to a cluster119864 which includes three straight line elements This systemhas the following three input attributes
(1) the standard deviation for 1205931 1205932 and 120593
3(see
Figure 7(d)) it is denoted by 1199091
(2) the difference between 1205932and 120593
3 it is denoted by 119909
2
(3) the difference between the two lengths of the elements1198901and 1198903 it is denoted by 119909
3
6 Advances in Fuzzy Systems
y = 1205831large (x1)y = 1205831small (x1)
y
10
0 10 30x1
(a) Membership functions for 1199091
y = 1205832large (x2)y = 1205832small (x2)
y
10
0 10 15x2
(b) Membership functions for 1199092
y = 1205833large (x3)y = 1205833small (x3)
y
10
0 10 20x3
(c) Membership functions for 1199093
y = 120583plau (x)y = 120583implau (x)
y
10
02 05 08x
0
(d) Membership functions for consequence
Figure 8Membership functions 120583119894small and 120583119894
large (119894 = 1 2 3) aremembership functions for the fuzzy sets small and large in Rule 119895 (119895 = 1 2 3)120583plau and 120583implau are membership functions for the fuzzy sets plausible and implausible in the consequence
e1
e2
e3
e4
(a)
e1
e2
e3
e4
12059311205932
1205933 1205934
(b)
Figure 9 Elements for cross
The following rules are the fuzzy if-then rules The member-ship functions are omitted
Rule 1 If 1199091is large 119909
2is small and 119909
3is small then it is
plausible that 119864 is an arrow
Rule 2 If 1199091is small then it is implausible that 119864 is an arrow
Rule 3 If 1199092is large then it is implausible that 119864 is an arrow
Rule 4 If 1199093is large then it is implausible that 119864 is an arrow
32 Cross Classification A cross consists of four straight lineelements (see Figure 9(a)) A cluster 119864 is applied to the fuzzyinference system for cross classification if 119864 includes four
straight line elements 1198901 1198902 1198903 and 119890
4 This fuzzy inference
system requires the following two input attributes(1) the standard deviation of 120593
1 1205932 1205933 and 120593
4(see
Figure 9(b)) it is denoted by 1199091
(2) the standard deviation for 1205931+ 1205932 1205932+ 1205933 1205933+ 1205934
and 1205931+ 1205934 it is denoted by 119909
2
The fuzzy if-then rules are given belowwhile themembershipfunctions are omitted
Rule 1 If 1199091is small and 119909
2is small then it is plausible that 119864
is a cross
Rule 2 If 1199091is large then it is implausible that 119864 is a cross
Rule 3 If 1199092is large then it is implausible that 119864 is a cross
If the similarity for a cluster obtained by the crossclassification is larger than a threshold value the cluster isclassified as a cross Similarly if the similarity from the arrowclassification exceeds a threshold value the cluster is classifiedas an arrow
33 Route and Railway Classification Route and railwayclassifications are conducted by the following way
(1) We first detect a sequence denoted by 119876 of clusterssuch that every two adjacent clusters include a com-mon straight line element (see Figure 10)
Advances in Fuzzy Systems 7
The principal line Arrowheads
e1 e2
(a) Route with three arrows the elements 1198901 and 1198902are common to the left and the right arrows
The principal line
e1 e2
Auxiliary lines
(b) Railway with three crosses the elements 1198901 and 1198902are common to the left and the right crosses
Figure 10 Route and railway symbols
Pi+1(xi+1 yi+1)
Pi(xi yi)
eiminus1
ei
120593iminus1
120593i
Piminus1(ximinus1 yiminus1)
Figure 11 Curvature for digital curve
(2) If the sequence 119876 consists of arrows then the routeclassification is executed to calculate the similaritydenoted by 119904
1 for the sequence119876 If 119904
1is larger than a
threshold value the sequence119876 is classified as a route(3) If the sequence119876 consists of crosses then the railway
classification is executed and we obtain the similaritydenoted by 119904
2 for the sequence 119876 If 119904
2exceeds a
threshold value the sequence 119876 is classified as arailway
Fuzzy inference systems are applied to compute thetwo similarities To calculate the attribute values for thefuzzy inference systems we detect the principal line andthe arrowhead (or the auxiliary lines) for the sequence 119876 asfollows If two straight line elements 119890
1and 119890
2 satisfy the
following two geometric characteristics it is plausible that 1198901
and 1198902are part of the same straight line
(1) The elements 1198901and 1198902are connected at an intersec-
tion 119875(2) The curvature at the intersection is small
Let119862 be a digital curve and let119875119894be a point on119862 A curvature
120581 of 119862 at 119875119894is then defined as the subtraction |120593
119894minus 120593119894minus1| (see
Figure 11) that is
120581 =1003816100381610038161003816120593119894 minus 120593119894minus1
1003816100381610038161003816 (4)
where 120593119894is the measure of an angle which is formed between
the 119909-axis and the line segment from 119875119894to 119875119894+1
and 120593119894minus1
isalso the measure of an angle which is formed between the 119909-axis and the line segment from 119875
119894minus1to 119875119894[24] Two adjacent
elements 119890119894minus1
and 119890119894are merged into a single straight line
segment if the curvature between 119890119894minus1
and 119890119894is less than a
threshold value After that we extract the principal line andthe arrowheads (or the auxiliary lines) for the sequence 119876
After extracting the principle line and the arrowhead (orthe auxiliary lines) for the sequence 119876 two fuzzy inferencesystems are applied to classify the sequence 119876 as a routeor railway The first fuzzy inference system is for routeclassification It has the following two input attributes
(1) the standard deviation for the lengths of elements inthe principal line it is denoted by 119909
1
(2) the standard deviation for lengths of elements in thearrowheads it is denoted by 119909
2
The fuzzy if-then rules for the first fuzzy inferencesystem are denoted below but the membership functions areomitted
Rule 1 If 1199091is small and 119909
2is small then it is plausible that119876
is a route
Rule 2 If 1199091is large then it is implausible that 119876 is a route
Rule 3 If 1199092is large then it is implausible that 119876 is a route
The second fuzzy inference system is for railway classifi-cation and the description for the second system is omittedbecause it is similar to the first system
4 SVG and Edel Documents Production
In our method a figure consists of elements whose shapesare as follows straight line circle circular arc and curveAs described in Section 22 the shape for an element isdetermined by the method of least squares If an element hasbeen classified as a curve then this element is expressed bya piecewise cubic Bezier curve the piecewise cubic Beziercurve is detected by an algorithmbased on themethodof leastsquares [25]
A Bezier curve 119876(119905) of degree 119899 is defined as
119876 (119905) =
119899
sum
119894=0
119881119894119861119899
119894(119905) (119905 isin [0 1]) (5)
where the 119881119894s are the control points and the 119861119899
119894s are the
Bernstein polynomials
119861119899
119894(119905) = (
119899
119894) 119905119894(1 minus 119905)
119899minus119894(119894 = 0 1 119899) (6)
8 Advances in Fuzzy Systems
Figure 12 A single cubic Bezier segment
0
2
4
6
8
0h lt 2h 2ndash5h 5ndash10h 10hlt
Figure 13 Experience in computers the vertical axis indicates thenumber of participants
Equation (5) is called a cubic Bezier curve when 119899 = 3Figure 12 shows an example of cubic Bezier curves the fourpoints119881
011988111198812 and119881
3are the control points the thick curve
is the cubic Bezier curve and the polygon expresses the oneconstructed by the four control points
SVG has the formats to express the four shapes andtherefore we can create an SVG document for a hand-drawnmap directly once we have detected the shapes for all theelements of the map On the other hand an Edel documentis a collection of embossed dots So it is easy to create anEdel document once we have detected the shapes for all theelements of a hand-drawn map
5 Experimental Results
This section describes the accuracy for our classificationsystem Five participants drew 15 maps using pencils andpapers These maps were then captured by using an imagescanner the resolution of the scannerwas set to 100 dpiThesedigital images were saved as 24-bits bitmap images The sizesof the images are in 1 169times850 pixelsWe havemeasured theaccuracy for our classification system using precision recalland f -measure
In the 15 maps there are 50 traffic signals 15 destinations15 departure places and 40 landmarks The classificationaccuracy for our system is summarized in Table 1 We canconclude that our system can produce the Edel and SVGdocuments almost correctly from hand-drawn maps
0
2
4
6
8
Freq
uent
ly
Som
etim
es
Occ
asio
nally
Seld
om
Nev
er
Figure 14 Experience in PPT the vertical axis indicates the numberof participants
Table 1 Experimental results are as follows 119873 is the number of allobjects for the symbol119879 is the number of objects correctly classifiedas the symbol 119862 is the number of objects classified as the symbol 119875is precision 119877 is recall and 119865 is 119865-measure
Symbol 119873 119879 119862 119875 () 119877 () 119865 ()Traffic signal 50 50 52 962 100 981Destination 15 15 15 100 100 100Departure place 15 14 14 100 933 965Landmark 40 38 38 100 950 974Cross 130 125 125 100 962 981Arrow 122 121 121 100 992 996Route 15 14 16 875 933 903Railway 28 28 28 100 100 100
6 Usability Evaluation for Our System
61 Method We have studied usability evaluation for oursystemThirteen participants 12 males and 1 female aged 20participated in this investigation all of them are the third-year university students To evaluate the usability for oursystem we selected two common methods for productionof tactile graphics one is the method to use the softwaresystem Edel which assists us to draw diagram images fortactile graphics produced by Braille embossers and anotherone is the method using swell papers In the second methoda diagram image is transferred to a swell paper by a printerand so forth A swell paper has been coated with thermallyfoamed microcapsules that respond to irradiation and causethe dark image on the paper to swell creating the tactilegraphic Microsoft PowerPoint (PPT) was selected to draw adiagram image on a swell paper because this software systemis commonly used in the universities in Japan
Before starting the experiment we asked the followingquestions Q1 Q2 and Q3 to all the participants and theresults for the questionsQ1 andQ3 are summarized in Figures13 and 14 We omit to show the result for the question Q2because all the participants have no experience in using EdelAll the participants are familiar with using computers andhave much experience in using PTT
Advances in Fuzzy Systems 9
(a) (b)
(c) (d)
(e) (f)
Figure 15 Maps for Sessions 1 and 2 (a) Edel map for Session 1 (b) PPT map for Session 1 (c) hand-drawn map for Session 1 (d) Edel mapfor Session 2 (e) PPT map for Session 2 and (f) hand-drawn map for Session 2
Q1 How long do you use computers every weekQ2 Have you ever created figures by using EdelQ3 Have you ever created figures by using PPT
We had two sessions in this investigation For the firstsession a participant created the map diagram shown inFigures 15(a)ndash15(c) the digital files were produced by usingEdel PPT and our system Note that in our system a userfirst draws a map manually using a pencil and paper thenconverts the hand-drawn map to a bitmap image using animage scanner and finally the bitmap image is transformed
into the two digital files the Edel and SVG documents Afterconducting all tasks the participant was asked to answer thefollowing two questions Q4 and Q5
Q4 Were you able to easily create the digital file(s) for thetactile map
Q5 Do you want to use this software system to create thedigital file for a tactile map
After the first session the participant had 5 minutes fortraining the software systems by himselfherself The secondsession followed this learning session In the second session
10 Advances in Fuzzy Systems
the participant conducted similar tasks to the first sessionbut the maps were those shown in Figures 15(d)ndash15(f) Aftercreating all digital files the participant was asked to answertwo questions Q4 and Q5 again and was also asked to answerthe following question Q6
Q6 Were you able to draw the map as you like
The orders of the software systems were determined ran-domly in both the first and the second sessions
62 Results We applied two-way ANOVA to the answersfor the 2 questions Q4 and Q5 The first factor denoted byFactor 1 is the difference in systems and the second factordenoted by Factor 2 is the difference in learning Factor 1includes three levels Edel PPT and our system and Factor2 includes two levels before learning and after learning Theresults from the two-way ANOVA for Q4 shows that therewas no significant interactions between Factor 1 and Factor2 (119865(2 72) = 092 119875 = 040) Furthermore there was asignificant difference for Factor 1 (119865(2 72) = 1926119875 lt 001)while no significant difference existed for Factor 2 (119865(1 72) =158 119875 = 021)The results for Q5 are similar to those for Q4That is there was no significant interaction between Factor1 and Factor 2 (119865(2 72) = 064 119875 = 053) there was asignificant difference for Factor 1 (119865(2 72) = 1400119875 lt 001)and therewas no significant difference for Factor 2 (119865(1 72) =132 119875 = 025)
We then conducted a multiple comparison test for theanswers of the questions Q4 and Q5 we selected Tukeyrsquoshonestly significant difference (HSD) criterion as themultiplecomparison test For Q4 there was a significant differencebetween Edel and our system (119875 lt 001) and there was alsoa significant difference between PPT and our system (119875 lt
001) however there was no significant difference betweenEdel and PPT (see Figure 16) Furthermore for Q5 therewas a significant difference between Edel and our system(119875 lt 001) and therewas also a significant difference betweenPPT and our system (119875 lt 005) however we observed amarginally significant difference between Edel and PPT (119875 lt01) (see Figure 17)
Lastly we applied a one-way ANOVA to the answersfor the question Q6 the factor of this one-way ANOVA isthe difference between systems The results are shown inFigure 18 and the results for the one-way ANOVA show thatthere was a marginally significant difference (119865(2 36) = 259119875 = 0089) We then applied Tukeyrsquos HSD and we observeda marginally significant difference between Edel and oursystem
As a result of the discussion above we can concludethat users feel that our method is easier to produce tactilemaps than the conventional twomethods Furthermore mapimages produced by our system are visually as fine as mapimages produced by Edel and PPT
7 Conclusions
In this paper a computer-aided system for automating pro-duction of tactile graphics from hand-drawn maps has been
0
2
4
6
Edel PPT Our System
Scor
e
Before LearningAfter Learning
lowastlowastlowastlowast
Figure 16 Results for Q4 the average and standard error for eachsystem (lowastlowast119875 lt 001)
0
1
2
3
4
Edel PPT Our System
Scor
e
lowastlowastlowast
Before LearningAfter Learning
Figure 17 Results for Q5 the average and standard error for eachsystem (lowastlowast119875 lt 001 lowast119875 lt 005)
developed The system includes the following four mainprocedures (1) preprocessing (2) segmentation (3) patternrecognition based on fuzzy inference and (4) SVG and Edeldocuments production The preprocessing is applied to aninitial hand-drawn map in order to remove noise and thena binary image is obtained In the segmentation procedurethe binary image is first skeletonized and then the skeletonis segmented into elements by eliminating intersections andcorner points Then a pattern recognition method is appliedto recognize symbols Lastly SVG and Edel documents arecreated to save the recognition results The usability for oursystem was evaluated by comparing our method with the twoconventional systems for creating tactile maps The resultsshow that our system is easier to create tactile maps than theother two systems
In our fuzzy inference systems triangular and trapezoidalmembership functions are applied to define the fuzzy setsdue to their simplicity The shapes of membership functionsinfluence the accuracy of recognition results How to choosean optimal membership function is our future work In thisstudy we do not assume a map includes character stringsHowever adding character strings to a map is very important
Advances in Fuzzy Systems 11
0
1
2
3
4
Edel PPT Our System
Scor
e
Figure 18 Results for Q6 the average and standard error for eachsystem
because it increases the comprehension for the map In thepresent work the types of symbols used in hand-drawnmaps are restricted Developing a system for automatingtranslation of hand-drawn maps with character string andvarious symbols is one of our future works
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This work was supported by the JSPS Grant-in-Aid forScientific Research (C) no 24501198 The authors thank DrKeisuke Ido for giving much advice to the experiments ofusability evaluation
References
[1] J A Miele S Landau and D Gilden ldquoTalking TMAPautomated generation of audio-tactile maps using Smith-Kettlewellrsquos TMAP softwarerdquo The British Journal of VisualImpairment vol 24 no 2 pp 93ndash100 2006
[2] T Watanabe T Yamaguchi K Watanabe et al ldquoDevelopmentand evaluation of a tactile map automated creation systemaccessible to blind personsrdquo IEICE Transactions on Informationand Systems vol J94-D no 10 pp 1652ndash1663 2011 (Japanese)
[3] TMACS httptmacsengniigata-uacjptmacs-dev[4] httptenpuchizugsigojpshokuchizu[5] S E Krufka and K E Barner ldquoAutomatic production of tactile
graphics from scalable vector graphicsrdquo in Proceedings of the 7thInternational ACM SIGACCESS Conference on Computers andAccessibility (ASSETS rsquo05) pp 166ndash172 October 2005
[6] R E Ladner M Y Ivory R Rao et al ldquoAutomating tactilegraphics translationrdquo in Proceedings of the 7th InternationalACMSIGACCESSConference onComputer andAccessibility pp150ndash157 New York NY USA October 2005
[7] Edel httpwww7abiglobenejp EDEL-plus[8] J D Eisenberg SVG Essentials OrsquoREILLY 2002[9] K Broelemann X Jiang and A Schwering ldquoAutomatic street
graph construction in sketchmapsrdquo inGraph-Based Representa-tions in Pattern Recognition X Jiang M Ferrer and A Torsello
Eds vol 6658 of Lecture Notes in Computer Science pp 275ndash284 Springer Berlin Germany 2011
[10] L B Kara and T F Stahovich ldquoAn image-based trainablesymbol recognizer for hand-drawn sketchesrdquo Computers ampGraphics vol 29 no 4 pp 501ndash517 2005
[11] B Edwards and V Chandran ldquoMachine recognition of hand-drawn circuit diagramsrdquo in Proceedings of the IEEE InterntionalConference onAcoustics Speech and Signal Processing vol 6 pp3618ndash3621 June 2000
[12] P Sala ldquoA recognition system for symbols of electronic com-ponents in hand-written circuit diagramsrdquo CSC 2515-MachineLearning Project Report 2004
[13] Y Qiao and M Yasuhara ldquoRecovering dynamic informationfrom static handwritten imagesrdquo in Proceedings of the 9th Inter-national Workshop on Frontiers in Handwriting Recognition pp118ndash123 jpn October 2004
[14] M Notowidigdo and R C Miller ldquoOff-line sketch interpreta-tionrdquo in Proceedings of the AAAI Fall Symposium on MakingPen-Based Interaction Intelligent and Natural pp 120ndash126Arlington Va USA April 2004
[15] R J Gonzalez and R E Woods Digital Image ProcessingPearson Education Upper Saddle River NJ USA 3rd edition2007
[16] N Otsu ldquoA threshold selection method from gray-level his-togramsrdquo IEEE Transactions on Systems Man and Cyberneticsvol 9 no 1 pp 62ndash66 1979
[17] C J Hilditch ldquoLinear skeletons from square cupboardsrdquo inMachine Intelligence Vol IV B Meltzer andDMichie Eds pp403ndash420 Elsevier New York NY USA 1969
[18] S Hermann and R Klette ldquoGlobal curvature estimation for cor-ner detectionrdquo Communication and Information TechnologyResearch Technical Report 171 2005
[19] N Nain V Laxmi B Bhadviya and A Gopal ldquoCorner detec-tion using difference chain code as curvaturerdquo in Proceedingsof the 3rd IEEE International Conference on Signal ImageTechnologies and Internet Based Systems (SITIS 07) pp 821ndash825Shanghai China December 2007
[20] B Kerautret J Lachaud and B Naegel ldquoCurvature basedcorner detector for discrete noisy and multi-scale contoursrdquoInternational Journal of Shape Modeling vol 14 no 2 pp 127ndash145 2008
[21] K Wall and P Danielsson ldquoA fast sequential method forpolygonal approximation of digitized curvesrdquoComputer VisionGraphics amp Image Processing vol 28 no 2 pp 220ndash227 1984
[22] D Chetverikov ldquoA simple and efficient algorithm for detectionof high curvature points in planar curvesrdquo inComputer Analysisof Images and Patterns vol 2756 of Lecture Notes in ComputerScience pp 746ndash753 Springer Berlin Germany 2003
[23] T Terano K Asai and M Sugeno Fuzzy Systems and ItsApplications Academic Press Boston Mass USA 2nd edition1992
[24] N Ono and R Takiyama ldquoOn calculations of curvature ofsampled curvesrdquo Technical Report of IEICE IE93-74 1993(Japanese)
[25] P J Schneider ldquoAn algorithm for automatically fitting digitizedcurvesrdquo in Graphics Gems A S Glassner Ed pp 612ndash625Academic Press 1990
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Advances in Fuzzy Systems 3
(a) Original image (b) Thresholded image by Otsursquos method
(c) Thresholded top-hat image
Figure 4 Using the top-hat transformation for shading correction the top-hat transformation corrects the effects of nonuniformillumination and as a result the fine binary image is created using Otsursquos method
in [10ndash14] cannot be used to recognize the hand-drawn mapdue to their limitations So a new method is proposed torecognize the hand-drawn map in this paper
In the remainder of this section preprocessing trafficsignal detection segmentation and shape classification areexplained
21 Preprocessing and Traffic Signal Detection The prepro-cessing consists of the following three procedures
(1) An original color image is converted to a grayscaleimage 119868 by the formula
119868 = 02989 sdot 119877 + 05866 sdot 119866 + 01145 sdot 119861 (1)
where 119877 119866 and 119861 are the intensities for red greenand blue of the original image
(2) The two noise reduction methods are applied to thegrayscale image
(a) The median filter with 3 times 3 window is firstapplied to eliminate salt-and-pepper noise
(b) a grayscale morphological top-hat transforma-tion [15] is applied to correct the effects ofnonuniform illumination Here a disk structur-ing element of radius 3 pixels is used for thetop-hat transformation The effect for a top-hattransformation is explained in Figure 4
(3) The grayscale image is converted to a binary imageusing Otsursquos method [16]
Mathematical morphology is applied to extract trafficsignals from the binary image First an erosion operatoris performed to the binary image for three times and adilation operator is then applied to the eroded image for threetimes The structuring element is a disk of radius 3 pixelsThe erosion operation eliminates thinner line segments butit does not remove traffic signals completely The dilationoperation then detects all the traffic signals By this detectionthe bullet for the destination symbol is also extracted
22 Segmentation and Shape Classification After detectingtraffic signals we segment objects into fragments The fourprocedures are introduced to divide objects (1) thinning (2)short branch removal (3) feature point detection and (4)dividing
(1) Hilditchrsquos thinning algorithm [17] is first applied to thebinary image and the skeleton image is obtained
(2) Short branch removal is as follows Every shortbranch of the skeleton is removed if its length isshorter than a threshold value
(3) Feature point detection is as follows Intersections andcorner points are detected in this stage A 3times3 slidingwindow is applied to the skeleton image in order todetect intersectionsThemethod for detecting cornerpoints is explained below
(4) Dividing is as follows By removing all intersectionsand corner points in the skeleton the skeleton isdivided into fragments It is characteristic that every
4 Advances in Fuzzy Systems
Ox
y
P1(x1 y1)
P2(x2 y2)
P3(x3 y3)
Figure 5 Piecewise linear approximation
fragment has two endpoints or no endpoint Frag-ments are called elements in this paper
Corner Point Detection There have been many studies fordetecting corner points in digital images [18ndash20] We intro-duce a new method to precisely detect corner points forhand-drawn figures This method first finds a piecewiselinear approximation (PL approximation for short) for adigital curve this approximation is detected by Wall andDanielssonrsquos algorithm [21] Let 119874 119875
1 and 119875
2be the origin
and the two different points on the 119909119910 plane (see Figure 5) Ifthe ratio of the area for the triangle119874119875
11198752to the length of the
line segment1198741198752exceeds a threshold value then the point 119875
1
is chosen as a point of the PL approximation if not the nextpoint 119875
2is examined Let us consider a digital curve denoted
by a sequence of the points119862 = 1198750 1198751 119875
119899minus1 TheWall and
Danielssonrsquos algorithm is then presented below
Step 1 Set 119894 larr 1 and choose the point 1198750as the first starting
point for the PL approximation
Step 2 Let 119874 be the point 119875119894minus1
and set 119890119894larr 0
Step 3 Set 119894 larr 119894 + 1 and calculate 119890119894and ℓ119894 Consider
119890119894larr997888 119890119894minus1
+ Δ119890119894
ℓ119894larr997888 radic119909
2
119894+ 1199102
119894
(2)
where Δ119890119894is the value 119909
119894minus1119910119894minus 119909119894119910119894minus1
which is the singed areafor the parallelogram spanned by the two vectors 997888997888997888997888rarr119874119875
119894minus1and
997888997888rarr119874119875119894
Step 4 If |119890119894| le 119905 sdot ℓ
119894holds then go to Step 3 if not choose
119875119894minus1
as a point of the PL approximation and then go to Step 2Here 119905 is a threshold value
The above algorithm cannot detect corner points cor-rectly However if a corner point exists it must be close toa point of the PL approximation So we improve the Walland Danielssonrsquos algorithm by exchanging Step 4 with thefollowing Step 41015840 so that the algorithm is able to preciselydetect corner points of a hand-drawn digital curve
So we introduce two procedures explained below andour corner detection procedure is the one that is obtained byexchanging Step 4 with the following Step 41015840
Step 41015840 If |119890119894| le 119905 sdot ℓ
119894holds then go to Step 3 if not the
point 119875119894minus1
is detected as a point of the PL approximation andperform the following process to test whether a corner pointexists near the point 119875
119894minus1
(1) Conduct the following procedure 1 whose input is119875119894minus1
If this procedure returns a point choose thispoint as a corner point of the digital curve119862 and thengo to Step 2
(2) Conduct the following procedure 2 whose input is119875119894minus1
If this procedure returns a point choose thispoint as a corner point of the digital curve119862 and thengo to Step 2
Procedure 1 aims to detect a sharp corner point Thisprocedure is performed in the following steps
Step 1 Let 119875119894minus119904
(119904 gt 1) be a point of the digital curve 119862satisfying the following two conditions
(1) 119875119894minus119904
has been chosen as a point of the PL approxima-tion
(2) 119875119894minus1
is the next PL approximation point to 119875119894minus119904
Step 2 For every point 119875119894minus119904+119896
of the digital curve 119862 between119875119894minus119904
and 119875119894minus1
(119896 = 1 2 119904 minus 1) calculate the distancedenoted by ℓ
119896 between 119875
119894minus119904and 119875
119894minus119904+119896 Then let 119871 be the
sequence ℓ1 ℓ2 ℓ
119904minus1
Step 3 For the sequence 119871 if the distances for119871 monotonously increase in the middle and themonotonously decrease to the end that is the inequalitiesℓ1lt ℓ2lt sdot sdot sdot lt ℓ
119902gt ℓ119902+1
gt sdot sdot sdot gt ℓ119904minus1
hold then the point119875119894minus119904+119902
is determined as a corner point of the digital curve 119862Next the procedure 2 is explained it is motivated by the
method for the literature [22] Remember that the point 119875119894minus1
has been chosen as a point of the PL approximation
Step 1 For every 119875119894minus119896
(119896 = 1 2 21) calculate thesharpness 119904(119875
119894minus119896) the procedure to calculate sharpness is
explained below
Step 2 For the sequence of the sharpnesses119904(119875119894minus1) 119904(119875119894minus2) 119904(119875
119894minus21) calculate the simple moving
averages smv(119875119894minus1) smv(119875
119894minus21)
Step 3 If the maximum value of smv(119875119894minus1) smv(119875
119894minus21)
exceeds a threshold value then the point 119875119894minus119896
which gives themaximum value is determined as a corner point of the digitalcurve 119862 if not we decide that no corner point exists near thepoint 119875
119894minus1
The calculation for the sharpness 119904(119875) for a point 119875 is asfollows We first detect two subsequences of points denotedby 119865 = 119875
1198961
119875119896119899
and 119861 = 1198751199051
119875119905119898
in the following way(see Figure 6) We visit the points forward from 119875 and select
Advances in Fuzzy Systems 5
Direction
Digital curve
B F
Pt119898+1
Pt119898
Pt1Pt2Pt3
P
Pk119899+1
Pk119899
Pk1Pk2Pk3
Figure 6 Sharpness
a point 119875119896119895
the distance between 119875119896119895
and 119875 is more than orequal to 2 pixels and less than or equal to 10 pixels 119875
1198961
is thefirst point satisfying this condition We select points until thecondition breaks119875
119896119899
is the last point satisfying the conditionThe subsequence 119861 is obtained similarly while visiting thepoints backward from 119875 The sharpness 119904(119875) is then definedas the following formula
119904 (119875) =1
119899119898
119899
sum
119903=1
119898
sum
119904=1
(the angle of ang119875119896119903
119875119875119905119904
) (3)
Next shape classification is explained Every element isclassified into one of the four shapes straight line circlecircular arc and curve This classification is done by themethod of least squares That is we minimize the sum ofthe squared Euclidean distances between the points of theelement and the corresponding points on the model Theelement is then classified as the shape for the model if thesum is smaller than a threshold value A curve is expressed bya piecewise cubic Bezier curve we will describe cubic Beziercurves in Section 4
3 Object Classification
In this section object classification is described First ofall a landmark object is classified by the following simpleprocedure if an object is isolated and closed and has morethan 3 corners the object is classified as a landmark Exceptfor landmark classification fuzzy inference is applied todesign the classification methods The fuzzy inference sys-tems introduced in this paper are constructed by Mamdanirsquosfuzzy inference method [23] but the minimum operator isexchanged with the product operator To detect route andrailway objects it is needed to find arrows and crosses Soelements are first grouped into a single cluster if they areconnected to the same intersection Every cluster is thenexamined if it is an arrow or a cross Note that an elementcan belong to different two clusters when the two endpointsof the element connect to the different intersections
31 Arrow Classification An arrow consists of four or threestraight line elements (see Figures 7(a) and 7(b)) We applytwo fuzzy inference systems to calculate the similarity forarrow The following description is the outline of the firstfuzzy inference system
e1
e2
e3
e4
(a)
e1
e2
e3
(b)
12059311205932
1205933 1205934
e1
e2
e3
e4
(c)
1205931
1205932
1205933
e1
e2
e3
(d)
Figure 7 Elements for arrows
(1) We first choose a cluster 119864 which includes fourstraight line elements denoted by 119890
1 1198902 1198903 and 119890
4
The four angles 1205931 1205932 1205933 and 120593
4are then measured
(see Figure 7(c))
(2) Next the following three attribute values are calcu-lated
(1) the standard deviation of 1205931 1205932 1205933 and 120593
4 it is
denoted by 1199091
(2) the average value of |1205931minus 1205934| and |120593
2minus 1205933| it is
denoted by 1199092
(3) the difference between the two lengths of theelements 119890
2and 1198904 it is denoted by 119909
3
(3) The fuzzy inference system is applied to 1199091 1199092 and
1199093 The fuzzy if-then rules are denoted below and
the membership functions are shown in Figure 8 Wethen have a value 119904 of [0 1] from the system 119904 impliesthe similarity for 119864
Rule 1 If 1199091is large 119909
2is small and 119909
3is small then it is
plausible that 119864 is an arrow
Rule 2 If 1199091is small then it is implausible that 119864 is an arrow
Rule 3 If 1199092is large then it is implausible that 119864 is an arrow
Rule 4 If 1199093is large then it is implausible that 119864 is an arrow
The second fuzzy inference system is applied to a cluster119864 which includes three straight line elements This systemhas the following three input attributes
(1) the standard deviation for 1205931 1205932 and 120593
3(see
Figure 7(d)) it is denoted by 1199091
(2) the difference between 1205932and 120593
3 it is denoted by 119909
2
(3) the difference between the two lengths of the elements1198901and 1198903 it is denoted by 119909
3
6 Advances in Fuzzy Systems
y = 1205831large (x1)y = 1205831small (x1)
y
10
0 10 30x1
(a) Membership functions for 1199091
y = 1205832large (x2)y = 1205832small (x2)
y
10
0 10 15x2
(b) Membership functions for 1199092
y = 1205833large (x3)y = 1205833small (x3)
y
10
0 10 20x3
(c) Membership functions for 1199093
y = 120583plau (x)y = 120583implau (x)
y
10
02 05 08x
0
(d) Membership functions for consequence
Figure 8Membership functions 120583119894small and 120583119894
large (119894 = 1 2 3) aremembership functions for the fuzzy sets small and large in Rule 119895 (119895 = 1 2 3)120583plau and 120583implau are membership functions for the fuzzy sets plausible and implausible in the consequence
e1
e2
e3
e4
(a)
e1
e2
e3
e4
12059311205932
1205933 1205934
(b)
Figure 9 Elements for cross
The following rules are the fuzzy if-then rules The member-ship functions are omitted
Rule 1 If 1199091is large 119909
2is small and 119909
3is small then it is
plausible that 119864 is an arrow
Rule 2 If 1199091is small then it is implausible that 119864 is an arrow
Rule 3 If 1199092is large then it is implausible that 119864 is an arrow
Rule 4 If 1199093is large then it is implausible that 119864 is an arrow
32 Cross Classification A cross consists of four straight lineelements (see Figure 9(a)) A cluster 119864 is applied to the fuzzyinference system for cross classification if 119864 includes four
straight line elements 1198901 1198902 1198903 and 119890
4 This fuzzy inference
system requires the following two input attributes(1) the standard deviation of 120593
1 1205932 1205933 and 120593
4(see
Figure 9(b)) it is denoted by 1199091
(2) the standard deviation for 1205931+ 1205932 1205932+ 1205933 1205933+ 1205934
and 1205931+ 1205934 it is denoted by 119909
2
The fuzzy if-then rules are given belowwhile themembershipfunctions are omitted
Rule 1 If 1199091is small and 119909
2is small then it is plausible that 119864
is a cross
Rule 2 If 1199091is large then it is implausible that 119864 is a cross
Rule 3 If 1199092is large then it is implausible that 119864 is a cross
If the similarity for a cluster obtained by the crossclassification is larger than a threshold value the cluster isclassified as a cross Similarly if the similarity from the arrowclassification exceeds a threshold value the cluster is classifiedas an arrow
33 Route and Railway Classification Route and railwayclassifications are conducted by the following way
(1) We first detect a sequence denoted by 119876 of clusterssuch that every two adjacent clusters include a com-mon straight line element (see Figure 10)
Advances in Fuzzy Systems 7
The principal line Arrowheads
e1 e2
(a) Route with three arrows the elements 1198901 and 1198902are common to the left and the right arrows
The principal line
e1 e2
Auxiliary lines
(b) Railway with three crosses the elements 1198901 and 1198902are common to the left and the right crosses
Figure 10 Route and railway symbols
Pi+1(xi+1 yi+1)
Pi(xi yi)
eiminus1
ei
120593iminus1
120593i
Piminus1(ximinus1 yiminus1)
Figure 11 Curvature for digital curve
(2) If the sequence 119876 consists of arrows then the routeclassification is executed to calculate the similaritydenoted by 119904
1 for the sequence119876 If 119904
1is larger than a
threshold value the sequence119876 is classified as a route(3) If the sequence119876 consists of crosses then the railway
classification is executed and we obtain the similaritydenoted by 119904
2 for the sequence 119876 If 119904
2exceeds a
threshold value the sequence 119876 is classified as arailway
Fuzzy inference systems are applied to compute thetwo similarities To calculate the attribute values for thefuzzy inference systems we detect the principal line andthe arrowhead (or the auxiliary lines) for the sequence 119876 asfollows If two straight line elements 119890
1and 119890
2 satisfy the
following two geometric characteristics it is plausible that 1198901
and 1198902are part of the same straight line
(1) The elements 1198901and 1198902are connected at an intersec-
tion 119875(2) The curvature at the intersection is small
Let119862 be a digital curve and let119875119894be a point on119862 A curvature
120581 of 119862 at 119875119894is then defined as the subtraction |120593
119894minus 120593119894minus1| (see
Figure 11) that is
120581 =1003816100381610038161003816120593119894 minus 120593119894minus1
1003816100381610038161003816 (4)
where 120593119894is the measure of an angle which is formed between
the 119909-axis and the line segment from 119875119894to 119875119894+1
and 120593119894minus1
isalso the measure of an angle which is formed between the 119909-axis and the line segment from 119875
119894minus1to 119875119894[24] Two adjacent
elements 119890119894minus1
and 119890119894are merged into a single straight line
segment if the curvature between 119890119894minus1
and 119890119894is less than a
threshold value After that we extract the principal line andthe arrowheads (or the auxiliary lines) for the sequence 119876
After extracting the principle line and the arrowhead (orthe auxiliary lines) for the sequence 119876 two fuzzy inferencesystems are applied to classify the sequence 119876 as a routeor railway The first fuzzy inference system is for routeclassification It has the following two input attributes
(1) the standard deviation for the lengths of elements inthe principal line it is denoted by 119909
1
(2) the standard deviation for lengths of elements in thearrowheads it is denoted by 119909
2
The fuzzy if-then rules for the first fuzzy inferencesystem are denoted below but the membership functions areomitted
Rule 1 If 1199091is small and 119909
2is small then it is plausible that119876
is a route
Rule 2 If 1199091is large then it is implausible that 119876 is a route
Rule 3 If 1199092is large then it is implausible that 119876 is a route
The second fuzzy inference system is for railway classifi-cation and the description for the second system is omittedbecause it is similar to the first system
4 SVG and Edel Documents Production
In our method a figure consists of elements whose shapesare as follows straight line circle circular arc and curveAs described in Section 22 the shape for an element isdetermined by the method of least squares If an element hasbeen classified as a curve then this element is expressed bya piecewise cubic Bezier curve the piecewise cubic Beziercurve is detected by an algorithmbased on themethodof leastsquares [25]
A Bezier curve 119876(119905) of degree 119899 is defined as
119876 (119905) =
119899
sum
119894=0
119881119894119861119899
119894(119905) (119905 isin [0 1]) (5)
where the 119881119894s are the control points and the 119861119899
119894s are the
Bernstein polynomials
119861119899
119894(119905) = (
119899
119894) 119905119894(1 minus 119905)
119899minus119894(119894 = 0 1 119899) (6)
8 Advances in Fuzzy Systems
Figure 12 A single cubic Bezier segment
0
2
4
6
8
0h lt 2h 2ndash5h 5ndash10h 10hlt
Figure 13 Experience in computers the vertical axis indicates thenumber of participants
Equation (5) is called a cubic Bezier curve when 119899 = 3Figure 12 shows an example of cubic Bezier curves the fourpoints119881
011988111198812 and119881
3are the control points the thick curve
is the cubic Bezier curve and the polygon expresses the oneconstructed by the four control points
SVG has the formats to express the four shapes andtherefore we can create an SVG document for a hand-drawnmap directly once we have detected the shapes for all theelements of the map On the other hand an Edel documentis a collection of embossed dots So it is easy to create anEdel document once we have detected the shapes for all theelements of a hand-drawn map
5 Experimental Results
This section describes the accuracy for our classificationsystem Five participants drew 15 maps using pencils andpapers These maps were then captured by using an imagescanner the resolution of the scannerwas set to 100 dpiThesedigital images were saved as 24-bits bitmap images The sizesof the images are in 1 169times850 pixelsWe havemeasured theaccuracy for our classification system using precision recalland f -measure
In the 15 maps there are 50 traffic signals 15 destinations15 departure places and 40 landmarks The classificationaccuracy for our system is summarized in Table 1 We canconclude that our system can produce the Edel and SVGdocuments almost correctly from hand-drawn maps
0
2
4
6
8
Freq
uent
ly
Som
etim
es
Occ
asio
nally
Seld
om
Nev
er
Figure 14 Experience in PPT the vertical axis indicates the numberof participants
Table 1 Experimental results are as follows 119873 is the number of allobjects for the symbol119879 is the number of objects correctly classifiedas the symbol 119862 is the number of objects classified as the symbol 119875is precision 119877 is recall and 119865 is 119865-measure
Symbol 119873 119879 119862 119875 () 119877 () 119865 ()Traffic signal 50 50 52 962 100 981Destination 15 15 15 100 100 100Departure place 15 14 14 100 933 965Landmark 40 38 38 100 950 974Cross 130 125 125 100 962 981Arrow 122 121 121 100 992 996Route 15 14 16 875 933 903Railway 28 28 28 100 100 100
6 Usability Evaluation for Our System
61 Method We have studied usability evaluation for oursystemThirteen participants 12 males and 1 female aged 20participated in this investigation all of them are the third-year university students To evaluate the usability for oursystem we selected two common methods for productionof tactile graphics one is the method to use the softwaresystem Edel which assists us to draw diagram images fortactile graphics produced by Braille embossers and anotherone is the method using swell papers In the second methoda diagram image is transferred to a swell paper by a printerand so forth A swell paper has been coated with thermallyfoamed microcapsules that respond to irradiation and causethe dark image on the paper to swell creating the tactilegraphic Microsoft PowerPoint (PPT) was selected to draw adiagram image on a swell paper because this software systemis commonly used in the universities in Japan
Before starting the experiment we asked the followingquestions Q1 Q2 and Q3 to all the participants and theresults for the questionsQ1 andQ3 are summarized in Figures13 and 14 We omit to show the result for the question Q2because all the participants have no experience in using EdelAll the participants are familiar with using computers andhave much experience in using PTT
Advances in Fuzzy Systems 9
(a) (b)
(c) (d)
(e) (f)
Figure 15 Maps for Sessions 1 and 2 (a) Edel map for Session 1 (b) PPT map for Session 1 (c) hand-drawn map for Session 1 (d) Edel mapfor Session 2 (e) PPT map for Session 2 and (f) hand-drawn map for Session 2
Q1 How long do you use computers every weekQ2 Have you ever created figures by using EdelQ3 Have you ever created figures by using PPT
We had two sessions in this investigation For the firstsession a participant created the map diagram shown inFigures 15(a)ndash15(c) the digital files were produced by usingEdel PPT and our system Note that in our system a userfirst draws a map manually using a pencil and paper thenconverts the hand-drawn map to a bitmap image using animage scanner and finally the bitmap image is transformed
into the two digital files the Edel and SVG documents Afterconducting all tasks the participant was asked to answer thefollowing two questions Q4 and Q5
Q4 Were you able to easily create the digital file(s) for thetactile map
Q5 Do you want to use this software system to create thedigital file for a tactile map
After the first session the participant had 5 minutes fortraining the software systems by himselfherself The secondsession followed this learning session In the second session
10 Advances in Fuzzy Systems
the participant conducted similar tasks to the first sessionbut the maps were those shown in Figures 15(d)ndash15(f) Aftercreating all digital files the participant was asked to answertwo questions Q4 and Q5 again and was also asked to answerthe following question Q6
Q6 Were you able to draw the map as you like
The orders of the software systems were determined ran-domly in both the first and the second sessions
62 Results We applied two-way ANOVA to the answersfor the 2 questions Q4 and Q5 The first factor denoted byFactor 1 is the difference in systems and the second factordenoted by Factor 2 is the difference in learning Factor 1includes three levels Edel PPT and our system and Factor2 includes two levels before learning and after learning Theresults from the two-way ANOVA for Q4 shows that therewas no significant interactions between Factor 1 and Factor2 (119865(2 72) = 092 119875 = 040) Furthermore there was asignificant difference for Factor 1 (119865(2 72) = 1926119875 lt 001)while no significant difference existed for Factor 2 (119865(1 72) =158 119875 = 021)The results for Q5 are similar to those for Q4That is there was no significant interaction between Factor1 and Factor 2 (119865(2 72) = 064 119875 = 053) there was asignificant difference for Factor 1 (119865(2 72) = 1400119875 lt 001)and therewas no significant difference for Factor 2 (119865(1 72) =132 119875 = 025)
We then conducted a multiple comparison test for theanswers of the questions Q4 and Q5 we selected Tukeyrsquoshonestly significant difference (HSD) criterion as themultiplecomparison test For Q4 there was a significant differencebetween Edel and our system (119875 lt 001) and there was alsoa significant difference between PPT and our system (119875 lt
001) however there was no significant difference betweenEdel and PPT (see Figure 16) Furthermore for Q5 therewas a significant difference between Edel and our system(119875 lt 001) and therewas also a significant difference betweenPPT and our system (119875 lt 005) however we observed amarginally significant difference between Edel and PPT (119875 lt01) (see Figure 17)
Lastly we applied a one-way ANOVA to the answersfor the question Q6 the factor of this one-way ANOVA isthe difference between systems The results are shown inFigure 18 and the results for the one-way ANOVA show thatthere was a marginally significant difference (119865(2 36) = 259119875 = 0089) We then applied Tukeyrsquos HSD and we observeda marginally significant difference between Edel and oursystem
As a result of the discussion above we can concludethat users feel that our method is easier to produce tactilemaps than the conventional twomethods Furthermore mapimages produced by our system are visually as fine as mapimages produced by Edel and PPT
7 Conclusions
In this paper a computer-aided system for automating pro-duction of tactile graphics from hand-drawn maps has been
0
2
4
6
Edel PPT Our System
Scor
e
Before LearningAfter Learning
lowastlowastlowastlowast
Figure 16 Results for Q4 the average and standard error for eachsystem (lowastlowast119875 lt 001)
0
1
2
3
4
Edel PPT Our System
Scor
e
lowastlowastlowast
Before LearningAfter Learning
Figure 17 Results for Q5 the average and standard error for eachsystem (lowastlowast119875 lt 001 lowast119875 lt 005)
developed The system includes the following four mainprocedures (1) preprocessing (2) segmentation (3) patternrecognition based on fuzzy inference and (4) SVG and Edeldocuments production The preprocessing is applied to aninitial hand-drawn map in order to remove noise and thena binary image is obtained In the segmentation procedurethe binary image is first skeletonized and then the skeletonis segmented into elements by eliminating intersections andcorner points Then a pattern recognition method is appliedto recognize symbols Lastly SVG and Edel documents arecreated to save the recognition results The usability for oursystem was evaluated by comparing our method with the twoconventional systems for creating tactile maps The resultsshow that our system is easier to create tactile maps than theother two systems
In our fuzzy inference systems triangular and trapezoidalmembership functions are applied to define the fuzzy setsdue to their simplicity The shapes of membership functionsinfluence the accuracy of recognition results How to choosean optimal membership function is our future work In thisstudy we do not assume a map includes character stringsHowever adding character strings to a map is very important
Advances in Fuzzy Systems 11
0
1
2
3
4
Edel PPT Our System
Scor
e
Figure 18 Results for Q6 the average and standard error for eachsystem
because it increases the comprehension for the map In thepresent work the types of symbols used in hand-drawnmaps are restricted Developing a system for automatingtranslation of hand-drawn maps with character string andvarious symbols is one of our future works
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This work was supported by the JSPS Grant-in-Aid forScientific Research (C) no 24501198 The authors thank DrKeisuke Ido for giving much advice to the experiments ofusability evaluation
References
[1] J A Miele S Landau and D Gilden ldquoTalking TMAPautomated generation of audio-tactile maps using Smith-Kettlewellrsquos TMAP softwarerdquo The British Journal of VisualImpairment vol 24 no 2 pp 93ndash100 2006
[2] T Watanabe T Yamaguchi K Watanabe et al ldquoDevelopmentand evaluation of a tactile map automated creation systemaccessible to blind personsrdquo IEICE Transactions on Informationand Systems vol J94-D no 10 pp 1652ndash1663 2011 (Japanese)
[3] TMACS httptmacsengniigata-uacjptmacs-dev[4] httptenpuchizugsigojpshokuchizu[5] S E Krufka and K E Barner ldquoAutomatic production of tactile
graphics from scalable vector graphicsrdquo in Proceedings of the 7thInternational ACM SIGACCESS Conference on Computers andAccessibility (ASSETS rsquo05) pp 166ndash172 October 2005
[6] R E Ladner M Y Ivory R Rao et al ldquoAutomating tactilegraphics translationrdquo in Proceedings of the 7th InternationalACMSIGACCESSConference onComputer andAccessibility pp150ndash157 New York NY USA October 2005
[7] Edel httpwww7abiglobenejp EDEL-plus[8] J D Eisenberg SVG Essentials OrsquoREILLY 2002[9] K Broelemann X Jiang and A Schwering ldquoAutomatic street
graph construction in sketchmapsrdquo inGraph-Based Representa-tions in Pattern Recognition X Jiang M Ferrer and A Torsello
Eds vol 6658 of Lecture Notes in Computer Science pp 275ndash284 Springer Berlin Germany 2011
[10] L B Kara and T F Stahovich ldquoAn image-based trainablesymbol recognizer for hand-drawn sketchesrdquo Computers ampGraphics vol 29 no 4 pp 501ndash517 2005
[11] B Edwards and V Chandran ldquoMachine recognition of hand-drawn circuit diagramsrdquo in Proceedings of the IEEE InterntionalConference onAcoustics Speech and Signal Processing vol 6 pp3618ndash3621 June 2000
[12] P Sala ldquoA recognition system for symbols of electronic com-ponents in hand-written circuit diagramsrdquo CSC 2515-MachineLearning Project Report 2004
[13] Y Qiao and M Yasuhara ldquoRecovering dynamic informationfrom static handwritten imagesrdquo in Proceedings of the 9th Inter-national Workshop on Frontiers in Handwriting Recognition pp118ndash123 jpn October 2004
[14] M Notowidigdo and R C Miller ldquoOff-line sketch interpreta-tionrdquo in Proceedings of the AAAI Fall Symposium on MakingPen-Based Interaction Intelligent and Natural pp 120ndash126Arlington Va USA April 2004
[15] R J Gonzalez and R E Woods Digital Image ProcessingPearson Education Upper Saddle River NJ USA 3rd edition2007
[16] N Otsu ldquoA threshold selection method from gray-level his-togramsrdquo IEEE Transactions on Systems Man and Cyberneticsvol 9 no 1 pp 62ndash66 1979
[17] C J Hilditch ldquoLinear skeletons from square cupboardsrdquo inMachine Intelligence Vol IV B Meltzer andDMichie Eds pp403ndash420 Elsevier New York NY USA 1969
[18] S Hermann and R Klette ldquoGlobal curvature estimation for cor-ner detectionrdquo Communication and Information TechnologyResearch Technical Report 171 2005
[19] N Nain V Laxmi B Bhadviya and A Gopal ldquoCorner detec-tion using difference chain code as curvaturerdquo in Proceedingsof the 3rd IEEE International Conference on Signal ImageTechnologies and Internet Based Systems (SITIS 07) pp 821ndash825Shanghai China December 2007
[20] B Kerautret J Lachaud and B Naegel ldquoCurvature basedcorner detector for discrete noisy and multi-scale contoursrdquoInternational Journal of Shape Modeling vol 14 no 2 pp 127ndash145 2008
[21] K Wall and P Danielsson ldquoA fast sequential method forpolygonal approximation of digitized curvesrdquoComputer VisionGraphics amp Image Processing vol 28 no 2 pp 220ndash227 1984
[22] D Chetverikov ldquoA simple and efficient algorithm for detectionof high curvature points in planar curvesrdquo inComputer Analysisof Images and Patterns vol 2756 of Lecture Notes in ComputerScience pp 746ndash753 Springer Berlin Germany 2003
[23] T Terano K Asai and M Sugeno Fuzzy Systems and ItsApplications Academic Press Boston Mass USA 2nd edition1992
[24] N Ono and R Takiyama ldquoOn calculations of curvature ofsampled curvesrdquo Technical Report of IEICE IE93-74 1993(Japanese)
[25] P J Schneider ldquoAn algorithm for automatically fitting digitizedcurvesrdquo in Graphics Gems A S Glassner Ed pp 612ndash625Academic Press 1990
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
4 Advances in Fuzzy Systems
Ox
y
P1(x1 y1)
P2(x2 y2)
P3(x3 y3)
Figure 5 Piecewise linear approximation
fragment has two endpoints or no endpoint Frag-ments are called elements in this paper
Corner Point Detection There have been many studies fordetecting corner points in digital images [18ndash20] We intro-duce a new method to precisely detect corner points forhand-drawn figures This method first finds a piecewiselinear approximation (PL approximation for short) for adigital curve this approximation is detected by Wall andDanielssonrsquos algorithm [21] Let 119874 119875
1 and 119875
2be the origin
and the two different points on the 119909119910 plane (see Figure 5) Ifthe ratio of the area for the triangle119874119875
11198752to the length of the
line segment1198741198752exceeds a threshold value then the point 119875
1
is chosen as a point of the PL approximation if not the nextpoint 119875
2is examined Let us consider a digital curve denoted
by a sequence of the points119862 = 1198750 1198751 119875
119899minus1 TheWall and
Danielssonrsquos algorithm is then presented below
Step 1 Set 119894 larr 1 and choose the point 1198750as the first starting
point for the PL approximation
Step 2 Let 119874 be the point 119875119894minus1
and set 119890119894larr 0
Step 3 Set 119894 larr 119894 + 1 and calculate 119890119894and ℓ119894 Consider
119890119894larr997888 119890119894minus1
+ Δ119890119894
ℓ119894larr997888 radic119909
2
119894+ 1199102
119894
(2)
where Δ119890119894is the value 119909
119894minus1119910119894minus 119909119894119910119894minus1
which is the singed areafor the parallelogram spanned by the two vectors 997888997888997888997888rarr119874119875
119894minus1and
997888997888rarr119874119875119894
Step 4 If |119890119894| le 119905 sdot ℓ
119894holds then go to Step 3 if not choose
119875119894minus1
as a point of the PL approximation and then go to Step 2Here 119905 is a threshold value
The above algorithm cannot detect corner points cor-rectly However if a corner point exists it must be close toa point of the PL approximation So we improve the Walland Danielssonrsquos algorithm by exchanging Step 4 with thefollowing Step 41015840 so that the algorithm is able to preciselydetect corner points of a hand-drawn digital curve
So we introduce two procedures explained below andour corner detection procedure is the one that is obtained byexchanging Step 4 with the following Step 41015840
Step 41015840 If |119890119894| le 119905 sdot ℓ
119894holds then go to Step 3 if not the
point 119875119894minus1
is detected as a point of the PL approximation andperform the following process to test whether a corner pointexists near the point 119875
119894minus1
(1) Conduct the following procedure 1 whose input is119875119894minus1
If this procedure returns a point choose thispoint as a corner point of the digital curve119862 and thengo to Step 2
(2) Conduct the following procedure 2 whose input is119875119894minus1
If this procedure returns a point choose thispoint as a corner point of the digital curve119862 and thengo to Step 2
Procedure 1 aims to detect a sharp corner point Thisprocedure is performed in the following steps
Step 1 Let 119875119894minus119904
(119904 gt 1) be a point of the digital curve 119862satisfying the following two conditions
(1) 119875119894minus119904
has been chosen as a point of the PL approxima-tion
(2) 119875119894minus1
is the next PL approximation point to 119875119894minus119904
Step 2 For every point 119875119894minus119904+119896
of the digital curve 119862 between119875119894minus119904
and 119875119894minus1
(119896 = 1 2 119904 minus 1) calculate the distancedenoted by ℓ
119896 between 119875
119894minus119904and 119875
119894minus119904+119896 Then let 119871 be the
sequence ℓ1 ℓ2 ℓ
119904minus1
Step 3 For the sequence 119871 if the distances for119871 monotonously increase in the middle and themonotonously decrease to the end that is the inequalitiesℓ1lt ℓ2lt sdot sdot sdot lt ℓ
119902gt ℓ119902+1
gt sdot sdot sdot gt ℓ119904minus1
hold then the point119875119894minus119904+119902
is determined as a corner point of the digital curve 119862Next the procedure 2 is explained it is motivated by the
method for the literature [22] Remember that the point 119875119894minus1
has been chosen as a point of the PL approximation
Step 1 For every 119875119894minus119896
(119896 = 1 2 21) calculate thesharpness 119904(119875
119894minus119896) the procedure to calculate sharpness is
explained below
Step 2 For the sequence of the sharpnesses119904(119875119894minus1) 119904(119875119894minus2) 119904(119875
119894minus21) calculate the simple moving
averages smv(119875119894minus1) smv(119875
119894minus21)
Step 3 If the maximum value of smv(119875119894minus1) smv(119875
119894minus21)
exceeds a threshold value then the point 119875119894minus119896
which gives themaximum value is determined as a corner point of the digitalcurve 119862 if not we decide that no corner point exists near thepoint 119875
119894minus1
The calculation for the sharpness 119904(119875) for a point 119875 is asfollows We first detect two subsequences of points denotedby 119865 = 119875
1198961
119875119896119899
and 119861 = 1198751199051
119875119905119898
in the following way(see Figure 6) We visit the points forward from 119875 and select
Advances in Fuzzy Systems 5
Direction
Digital curve
B F
Pt119898+1
Pt119898
Pt1Pt2Pt3
P
Pk119899+1
Pk119899
Pk1Pk2Pk3
Figure 6 Sharpness
a point 119875119896119895
the distance between 119875119896119895
and 119875 is more than orequal to 2 pixels and less than or equal to 10 pixels 119875
1198961
is thefirst point satisfying this condition We select points until thecondition breaks119875
119896119899
is the last point satisfying the conditionThe subsequence 119861 is obtained similarly while visiting thepoints backward from 119875 The sharpness 119904(119875) is then definedas the following formula
119904 (119875) =1
119899119898
119899
sum
119903=1
119898
sum
119904=1
(the angle of ang119875119896119903
119875119875119905119904
) (3)
Next shape classification is explained Every element isclassified into one of the four shapes straight line circlecircular arc and curve This classification is done by themethod of least squares That is we minimize the sum ofthe squared Euclidean distances between the points of theelement and the corresponding points on the model Theelement is then classified as the shape for the model if thesum is smaller than a threshold value A curve is expressed bya piecewise cubic Bezier curve we will describe cubic Beziercurves in Section 4
3 Object Classification
In this section object classification is described First ofall a landmark object is classified by the following simpleprocedure if an object is isolated and closed and has morethan 3 corners the object is classified as a landmark Exceptfor landmark classification fuzzy inference is applied todesign the classification methods The fuzzy inference sys-tems introduced in this paper are constructed by Mamdanirsquosfuzzy inference method [23] but the minimum operator isexchanged with the product operator To detect route andrailway objects it is needed to find arrows and crosses Soelements are first grouped into a single cluster if they areconnected to the same intersection Every cluster is thenexamined if it is an arrow or a cross Note that an elementcan belong to different two clusters when the two endpointsof the element connect to the different intersections
31 Arrow Classification An arrow consists of four or threestraight line elements (see Figures 7(a) and 7(b)) We applytwo fuzzy inference systems to calculate the similarity forarrow The following description is the outline of the firstfuzzy inference system
e1
e2
e3
e4
(a)
e1
e2
e3
(b)
12059311205932
1205933 1205934
e1
e2
e3
e4
(c)
1205931
1205932
1205933
e1
e2
e3
(d)
Figure 7 Elements for arrows
(1) We first choose a cluster 119864 which includes fourstraight line elements denoted by 119890
1 1198902 1198903 and 119890
4
The four angles 1205931 1205932 1205933 and 120593
4are then measured
(see Figure 7(c))
(2) Next the following three attribute values are calcu-lated
(1) the standard deviation of 1205931 1205932 1205933 and 120593
4 it is
denoted by 1199091
(2) the average value of |1205931minus 1205934| and |120593
2minus 1205933| it is
denoted by 1199092
(3) the difference between the two lengths of theelements 119890
2and 1198904 it is denoted by 119909
3
(3) The fuzzy inference system is applied to 1199091 1199092 and
1199093 The fuzzy if-then rules are denoted below and
the membership functions are shown in Figure 8 Wethen have a value 119904 of [0 1] from the system 119904 impliesthe similarity for 119864
Rule 1 If 1199091is large 119909
2is small and 119909
3is small then it is
plausible that 119864 is an arrow
Rule 2 If 1199091is small then it is implausible that 119864 is an arrow
Rule 3 If 1199092is large then it is implausible that 119864 is an arrow
Rule 4 If 1199093is large then it is implausible that 119864 is an arrow
The second fuzzy inference system is applied to a cluster119864 which includes three straight line elements This systemhas the following three input attributes
(1) the standard deviation for 1205931 1205932 and 120593
3(see
Figure 7(d)) it is denoted by 1199091
(2) the difference between 1205932and 120593
3 it is denoted by 119909
2
(3) the difference between the two lengths of the elements1198901and 1198903 it is denoted by 119909
3
6 Advances in Fuzzy Systems
y = 1205831large (x1)y = 1205831small (x1)
y
10
0 10 30x1
(a) Membership functions for 1199091
y = 1205832large (x2)y = 1205832small (x2)
y
10
0 10 15x2
(b) Membership functions for 1199092
y = 1205833large (x3)y = 1205833small (x3)
y
10
0 10 20x3
(c) Membership functions for 1199093
y = 120583plau (x)y = 120583implau (x)
y
10
02 05 08x
0
(d) Membership functions for consequence
Figure 8Membership functions 120583119894small and 120583119894
large (119894 = 1 2 3) aremembership functions for the fuzzy sets small and large in Rule 119895 (119895 = 1 2 3)120583plau and 120583implau are membership functions for the fuzzy sets plausible and implausible in the consequence
e1
e2
e3
e4
(a)
e1
e2
e3
e4
12059311205932
1205933 1205934
(b)
Figure 9 Elements for cross
The following rules are the fuzzy if-then rules The member-ship functions are omitted
Rule 1 If 1199091is large 119909
2is small and 119909
3is small then it is
plausible that 119864 is an arrow
Rule 2 If 1199091is small then it is implausible that 119864 is an arrow
Rule 3 If 1199092is large then it is implausible that 119864 is an arrow
Rule 4 If 1199093is large then it is implausible that 119864 is an arrow
32 Cross Classification A cross consists of four straight lineelements (see Figure 9(a)) A cluster 119864 is applied to the fuzzyinference system for cross classification if 119864 includes four
straight line elements 1198901 1198902 1198903 and 119890
4 This fuzzy inference
system requires the following two input attributes(1) the standard deviation of 120593
1 1205932 1205933 and 120593
4(see
Figure 9(b)) it is denoted by 1199091
(2) the standard deviation for 1205931+ 1205932 1205932+ 1205933 1205933+ 1205934
and 1205931+ 1205934 it is denoted by 119909
2
The fuzzy if-then rules are given belowwhile themembershipfunctions are omitted
Rule 1 If 1199091is small and 119909
2is small then it is plausible that 119864
is a cross
Rule 2 If 1199091is large then it is implausible that 119864 is a cross
Rule 3 If 1199092is large then it is implausible that 119864 is a cross
If the similarity for a cluster obtained by the crossclassification is larger than a threshold value the cluster isclassified as a cross Similarly if the similarity from the arrowclassification exceeds a threshold value the cluster is classifiedas an arrow
33 Route and Railway Classification Route and railwayclassifications are conducted by the following way
(1) We first detect a sequence denoted by 119876 of clusterssuch that every two adjacent clusters include a com-mon straight line element (see Figure 10)
Advances in Fuzzy Systems 7
The principal line Arrowheads
e1 e2
(a) Route with three arrows the elements 1198901 and 1198902are common to the left and the right arrows
The principal line
e1 e2
Auxiliary lines
(b) Railway with three crosses the elements 1198901 and 1198902are common to the left and the right crosses
Figure 10 Route and railway symbols
Pi+1(xi+1 yi+1)
Pi(xi yi)
eiminus1
ei
120593iminus1
120593i
Piminus1(ximinus1 yiminus1)
Figure 11 Curvature for digital curve
(2) If the sequence 119876 consists of arrows then the routeclassification is executed to calculate the similaritydenoted by 119904
1 for the sequence119876 If 119904
1is larger than a
threshold value the sequence119876 is classified as a route(3) If the sequence119876 consists of crosses then the railway
classification is executed and we obtain the similaritydenoted by 119904
2 for the sequence 119876 If 119904
2exceeds a
threshold value the sequence 119876 is classified as arailway
Fuzzy inference systems are applied to compute thetwo similarities To calculate the attribute values for thefuzzy inference systems we detect the principal line andthe arrowhead (or the auxiliary lines) for the sequence 119876 asfollows If two straight line elements 119890
1and 119890
2 satisfy the
following two geometric characteristics it is plausible that 1198901
and 1198902are part of the same straight line
(1) The elements 1198901and 1198902are connected at an intersec-
tion 119875(2) The curvature at the intersection is small
Let119862 be a digital curve and let119875119894be a point on119862 A curvature
120581 of 119862 at 119875119894is then defined as the subtraction |120593
119894minus 120593119894minus1| (see
Figure 11) that is
120581 =1003816100381610038161003816120593119894 minus 120593119894minus1
1003816100381610038161003816 (4)
where 120593119894is the measure of an angle which is formed between
the 119909-axis and the line segment from 119875119894to 119875119894+1
and 120593119894minus1
isalso the measure of an angle which is formed between the 119909-axis and the line segment from 119875
119894minus1to 119875119894[24] Two adjacent
elements 119890119894minus1
and 119890119894are merged into a single straight line
segment if the curvature between 119890119894minus1
and 119890119894is less than a
threshold value After that we extract the principal line andthe arrowheads (or the auxiliary lines) for the sequence 119876
After extracting the principle line and the arrowhead (orthe auxiliary lines) for the sequence 119876 two fuzzy inferencesystems are applied to classify the sequence 119876 as a routeor railway The first fuzzy inference system is for routeclassification It has the following two input attributes
(1) the standard deviation for the lengths of elements inthe principal line it is denoted by 119909
1
(2) the standard deviation for lengths of elements in thearrowheads it is denoted by 119909
2
The fuzzy if-then rules for the first fuzzy inferencesystem are denoted below but the membership functions areomitted
Rule 1 If 1199091is small and 119909
2is small then it is plausible that119876
is a route
Rule 2 If 1199091is large then it is implausible that 119876 is a route
Rule 3 If 1199092is large then it is implausible that 119876 is a route
The second fuzzy inference system is for railway classifi-cation and the description for the second system is omittedbecause it is similar to the first system
4 SVG and Edel Documents Production
In our method a figure consists of elements whose shapesare as follows straight line circle circular arc and curveAs described in Section 22 the shape for an element isdetermined by the method of least squares If an element hasbeen classified as a curve then this element is expressed bya piecewise cubic Bezier curve the piecewise cubic Beziercurve is detected by an algorithmbased on themethodof leastsquares [25]
A Bezier curve 119876(119905) of degree 119899 is defined as
119876 (119905) =
119899
sum
119894=0
119881119894119861119899
119894(119905) (119905 isin [0 1]) (5)
where the 119881119894s are the control points and the 119861119899
119894s are the
Bernstein polynomials
119861119899
119894(119905) = (
119899
119894) 119905119894(1 minus 119905)
119899minus119894(119894 = 0 1 119899) (6)
8 Advances in Fuzzy Systems
Figure 12 A single cubic Bezier segment
0
2
4
6
8
0h lt 2h 2ndash5h 5ndash10h 10hlt
Figure 13 Experience in computers the vertical axis indicates thenumber of participants
Equation (5) is called a cubic Bezier curve when 119899 = 3Figure 12 shows an example of cubic Bezier curves the fourpoints119881
011988111198812 and119881
3are the control points the thick curve
is the cubic Bezier curve and the polygon expresses the oneconstructed by the four control points
SVG has the formats to express the four shapes andtherefore we can create an SVG document for a hand-drawnmap directly once we have detected the shapes for all theelements of the map On the other hand an Edel documentis a collection of embossed dots So it is easy to create anEdel document once we have detected the shapes for all theelements of a hand-drawn map
5 Experimental Results
This section describes the accuracy for our classificationsystem Five participants drew 15 maps using pencils andpapers These maps were then captured by using an imagescanner the resolution of the scannerwas set to 100 dpiThesedigital images were saved as 24-bits bitmap images The sizesof the images are in 1 169times850 pixelsWe havemeasured theaccuracy for our classification system using precision recalland f -measure
In the 15 maps there are 50 traffic signals 15 destinations15 departure places and 40 landmarks The classificationaccuracy for our system is summarized in Table 1 We canconclude that our system can produce the Edel and SVGdocuments almost correctly from hand-drawn maps
0
2
4
6
8
Freq
uent
ly
Som
etim
es
Occ
asio
nally
Seld
om
Nev
er
Figure 14 Experience in PPT the vertical axis indicates the numberof participants
Table 1 Experimental results are as follows 119873 is the number of allobjects for the symbol119879 is the number of objects correctly classifiedas the symbol 119862 is the number of objects classified as the symbol 119875is precision 119877 is recall and 119865 is 119865-measure
Symbol 119873 119879 119862 119875 () 119877 () 119865 ()Traffic signal 50 50 52 962 100 981Destination 15 15 15 100 100 100Departure place 15 14 14 100 933 965Landmark 40 38 38 100 950 974Cross 130 125 125 100 962 981Arrow 122 121 121 100 992 996Route 15 14 16 875 933 903Railway 28 28 28 100 100 100
6 Usability Evaluation for Our System
61 Method We have studied usability evaluation for oursystemThirteen participants 12 males and 1 female aged 20participated in this investigation all of them are the third-year university students To evaluate the usability for oursystem we selected two common methods for productionof tactile graphics one is the method to use the softwaresystem Edel which assists us to draw diagram images fortactile graphics produced by Braille embossers and anotherone is the method using swell papers In the second methoda diagram image is transferred to a swell paper by a printerand so forth A swell paper has been coated with thermallyfoamed microcapsules that respond to irradiation and causethe dark image on the paper to swell creating the tactilegraphic Microsoft PowerPoint (PPT) was selected to draw adiagram image on a swell paper because this software systemis commonly used in the universities in Japan
Before starting the experiment we asked the followingquestions Q1 Q2 and Q3 to all the participants and theresults for the questionsQ1 andQ3 are summarized in Figures13 and 14 We omit to show the result for the question Q2because all the participants have no experience in using EdelAll the participants are familiar with using computers andhave much experience in using PTT
Advances in Fuzzy Systems 9
(a) (b)
(c) (d)
(e) (f)
Figure 15 Maps for Sessions 1 and 2 (a) Edel map for Session 1 (b) PPT map for Session 1 (c) hand-drawn map for Session 1 (d) Edel mapfor Session 2 (e) PPT map for Session 2 and (f) hand-drawn map for Session 2
Q1 How long do you use computers every weekQ2 Have you ever created figures by using EdelQ3 Have you ever created figures by using PPT
We had two sessions in this investigation For the firstsession a participant created the map diagram shown inFigures 15(a)ndash15(c) the digital files were produced by usingEdel PPT and our system Note that in our system a userfirst draws a map manually using a pencil and paper thenconverts the hand-drawn map to a bitmap image using animage scanner and finally the bitmap image is transformed
into the two digital files the Edel and SVG documents Afterconducting all tasks the participant was asked to answer thefollowing two questions Q4 and Q5
Q4 Were you able to easily create the digital file(s) for thetactile map
Q5 Do you want to use this software system to create thedigital file for a tactile map
After the first session the participant had 5 minutes fortraining the software systems by himselfherself The secondsession followed this learning session In the second session
10 Advances in Fuzzy Systems
the participant conducted similar tasks to the first sessionbut the maps were those shown in Figures 15(d)ndash15(f) Aftercreating all digital files the participant was asked to answertwo questions Q4 and Q5 again and was also asked to answerthe following question Q6
Q6 Were you able to draw the map as you like
The orders of the software systems were determined ran-domly in both the first and the second sessions
62 Results We applied two-way ANOVA to the answersfor the 2 questions Q4 and Q5 The first factor denoted byFactor 1 is the difference in systems and the second factordenoted by Factor 2 is the difference in learning Factor 1includes three levels Edel PPT and our system and Factor2 includes two levels before learning and after learning Theresults from the two-way ANOVA for Q4 shows that therewas no significant interactions between Factor 1 and Factor2 (119865(2 72) = 092 119875 = 040) Furthermore there was asignificant difference for Factor 1 (119865(2 72) = 1926119875 lt 001)while no significant difference existed for Factor 2 (119865(1 72) =158 119875 = 021)The results for Q5 are similar to those for Q4That is there was no significant interaction between Factor1 and Factor 2 (119865(2 72) = 064 119875 = 053) there was asignificant difference for Factor 1 (119865(2 72) = 1400119875 lt 001)and therewas no significant difference for Factor 2 (119865(1 72) =132 119875 = 025)
We then conducted a multiple comparison test for theanswers of the questions Q4 and Q5 we selected Tukeyrsquoshonestly significant difference (HSD) criterion as themultiplecomparison test For Q4 there was a significant differencebetween Edel and our system (119875 lt 001) and there was alsoa significant difference between PPT and our system (119875 lt
001) however there was no significant difference betweenEdel and PPT (see Figure 16) Furthermore for Q5 therewas a significant difference between Edel and our system(119875 lt 001) and therewas also a significant difference betweenPPT and our system (119875 lt 005) however we observed amarginally significant difference between Edel and PPT (119875 lt01) (see Figure 17)
Lastly we applied a one-way ANOVA to the answersfor the question Q6 the factor of this one-way ANOVA isthe difference between systems The results are shown inFigure 18 and the results for the one-way ANOVA show thatthere was a marginally significant difference (119865(2 36) = 259119875 = 0089) We then applied Tukeyrsquos HSD and we observeda marginally significant difference between Edel and oursystem
As a result of the discussion above we can concludethat users feel that our method is easier to produce tactilemaps than the conventional twomethods Furthermore mapimages produced by our system are visually as fine as mapimages produced by Edel and PPT
7 Conclusions
In this paper a computer-aided system for automating pro-duction of tactile graphics from hand-drawn maps has been
0
2
4
6
Edel PPT Our System
Scor
e
Before LearningAfter Learning
lowastlowastlowastlowast
Figure 16 Results for Q4 the average and standard error for eachsystem (lowastlowast119875 lt 001)
0
1
2
3
4
Edel PPT Our System
Scor
e
lowastlowastlowast
Before LearningAfter Learning
Figure 17 Results for Q5 the average and standard error for eachsystem (lowastlowast119875 lt 001 lowast119875 lt 005)
developed The system includes the following four mainprocedures (1) preprocessing (2) segmentation (3) patternrecognition based on fuzzy inference and (4) SVG and Edeldocuments production The preprocessing is applied to aninitial hand-drawn map in order to remove noise and thena binary image is obtained In the segmentation procedurethe binary image is first skeletonized and then the skeletonis segmented into elements by eliminating intersections andcorner points Then a pattern recognition method is appliedto recognize symbols Lastly SVG and Edel documents arecreated to save the recognition results The usability for oursystem was evaluated by comparing our method with the twoconventional systems for creating tactile maps The resultsshow that our system is easier to create tactile maps than theother two systems
In our fuzzy inference systems triangular and trapezoidalmembership functions are applied to define the fuzzy setsdue to their simplicity The shapes of membership functionsinfluence the accuracy of recognition results How to choosean optimal membership function is our future work In thisstudy we do not assume a map includes character stringsHowever adding character strings to a map is very important
Advances in Fuzzy Systems 11
0
1
2
3
4
Edel PPT Our System
Scor
e
Figure 18 Results for Q6 the average and standard error for eachsystem
because it increases the comprehension for the map In thepresent work the types of symbols used in hand-drawnmaps are restricted Developing a system for automatingtranslation of hand-drawn maps with character string andvarious symbols is one of our future works
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This work was supported by the JSPS Grant-in-Aid forScientific Research (C) no 24501198 The authors thank DrKeisuke Ido for giving much advice to the experiments ofusability evaluation
References
[1] J A Miele S Landau and D Gilden ldquoTalking TMAPautomated generation of audio-tactile maps using Smith-Kettlewellrsquos TMAP softwarerdquo The British Journal of VisualImpairment vol 24 no 2 pp 93ndash100 2006
[2] T Watanabe T Yamaguchi K Watanabe et al ldquoDevelopmentand evaluation of a tactile map automated creation systemaccessible to blind personsrdquo IEICE Transactions on Informationand Systems vol J94-D no 10 pp 1652ndash1663 2011 (Japanese)
[3] TMACS httptmacsengniigata-uacjptmacs-dev[4] httptenpuchizugsigojpshokuchizu[5] S E Krufka and K E Barner ldquoAutomatic production of tactile
graphics from scalable vector graphicsrdquo in Proceedings of the 7thInternational ACM SIGACCESS Conference on Computers andAccessibility (ASSETS rsquo05) pp 166ndash172 October 2005
[6] R E Ladner M Y Ivory R Rao et al ldquoAutomating tactilegraphics translationrdquo in Proceedings of the 7th InternationalACMSIGACCESSConference onComputer andAccessibility pp150ndash157 New York NY USA October 2005
[7] Edel httpwww7abiglobenejp EDEL-plus[8] J D Eisenberg SVG Essentials OrsquoREILLY 2002[9] K Broelemann X Jiang and A Schwering ldquoAutomatic street
graph construction in sketchmapsrdquo inGraph-Based Representa-tions in Pattern Recognition X Jiang M Ferrer and A Torsello
Eds vol 6658 of Lecture Notes in Computer Science pp 275ndash284 Springer Berlin Germany 2011
[10] L B Kara and T F Stahovich ldquoAn image-based trainablesymbol recognizer for hand-drawn sketchesrdquo Computers ampGraphics vol 29 no 4 pp 501ndash517 2005
[11] B Edwards and V Chandran ldquoMachine recognition of hand-drawn circuit diagramsrdquo in Proceedings of the IEEE InterntionalConference onAcoustics Speech and Signal Processing vol 6 pp3618ndash3621 June 2000
[12] P Sala ldquoA recognition system for symbols of electronic com-ponents in hand-written circuit diagramsrdquo CSC 2515-MachineLearning Project Report 2004
[13] Y Qiao and M Yasuhara ldquoRecovering dynamic informationfrom static handwritten imagesrdquo in Proceedings of the 9th Inter-national Workshop on Frontiers in Handwriting Recognition pp118ndash123 jpn October 2004
[14] M Notowidigdo and R C Miller ldquoOff-line sketch interpreta-tionrdquo in Proceedings of the AAAI Fall Symposium on MakingPen-Based Interaction Intelligent and Natural pp 120ndash126Arlington Va USA April 2004
[15] R J Gonzalez and R E Woods Digital Image ProcessingPearson Education Upper Saddle River NJ USA 3rd edition2007
[16] N Otsu ldquoA threshold selection method from gray-level his-togramsrdquo IEEE Transactions on Systems Man and Cyberneticsvol 9 no 1 pp 62ndash66 1979
[17] C J Hilditch ldquoLinear skeletons from square cupboardsrdquo inMachine Intelligence Vol IV B Meltzer andDMichie Eds pp403ndash420 Elsevier New York NY USA 1969
[18] S Hermann and R Klette ldquoGlobal curvature estimation for cor-ner detectionrdquo Communication and Information TechnologyResearch Technical Report 171 2005
[19] N Nain V Laxmi B Bhadviya and A Gopal ldquoCorner detec-tion using difference chain code as curvaturerdquo in Proceedingsof the 3rd IEEE International Conference on Signal ImageTechnologies and Internet Based Systems (SITIS 07) pp 821ndash825Shanghai China December 2007
[20] B Kerautret J Lachaud and B Naegel ldquoCurvature basedcorner detector for discrete noisy and multi-scale contoursrdquoInternational Journal of Shape Modeling vol 14 no 2 pp 127ndash145 2008
[21] K Wall and P Danielsson ldquoA fast sequential method forpolygonal approximation of digitized curvesrdquoComputer VisionGraphics amp Image Processing vol 28 no 2 pp 220ndash227 1984
[22] D Chetverikov ldquoA simple and efficient algorithm for detectionof high curvature points in planar curvesrdquo inComputer Analysisof Images and Patterns vol 2756 of Lecture Notes in ComputerScience pp 746ndash753 Springer Berlin Germany 2003
[23] T Terano K Asai and M Sugeno Fuzzy Systems and ItsApplications Academic Press Boston Mass USA 2nd edition1992
[24] N Ono and R Takiyama ldquoOn calculations of curvature ofsampled curvesrdquo Technical Report of IEICE IE93-74 1993(Japanese)
[25] P J Schneider ldquoAn algorithm for automatically fitting digitizedcurvesrdquo in Graphics Gems A S Glassner Ed pp 612ndash625Academic Press 1990
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Advances in Fuzzy Systems 5
Direction
Digital curve
B F
Pt119898+1
Pt119898
Pt1Pt2Pt3
P
Pk119899+1
Pk119899
Pk1Pk2Pk3
Figure 6 Sharpness
a point 119875119896119895
the distance between 119875119896119895
and 119875 is more than orequal to 2 pixels and less than or equal to 10 pixels 119875
1198961
is thefirst point satisfying this condition We select points until thecondition breaks119875
119896119899
is the last point satisfying the conditionThe subsequence 119861 is obtained similarly while visiting thepoints backward from 119875 The sharpness 119904(119875) is then definedas the following formula
119904 (119875) =1
119899119898
119899
sum
119903=1
119898
sum
119904=1
(the angle of ang119875119896119903
119875119875119905119904
) (3)
Next shape classification is explained Every element isclassified into one of the four shapes straight line circlecircular arc and curve This classification is done by themethod of least squares That is we minimize the sum ofthe squared Euclidean distances between the points of theelement and the corresponding points on the model Theelement is then classified as the shape for the model if thesum is smaller than a threshold value A curve is expressed bya piecewise cubic Bezier curve we will describe cubic Beziercurves in Section 4
3 Object Classification
In this section object classification is described First ofall a landmark object is classified by the following simpleprocedure if an object is isolated and closed and has morethan 3 corners the object is classified as a landmark Exceptfor landmark classification fuzzy inference is applied todesign the classification methods The fuzzy inference sys-tems introduced in this paper are constructed by Mamdanirsquosfuzzy inference method [23] but the minimum operator isexchanged with the product operator To detect route andrailway objects it is needed to find arrows and crosses Soelements are first grouped into a single cluster if they areconnected to the same intersection Every cluster is thenexamined if it is an arrow or a cross Note that an elementcan belong to different two clusters when the two endpointsof the element connect to the different intersections
31 Arrow Classification An arrow consists of four or threestraight line elements (see Figures 7(a) and 7(b)) We applytwo fuzzy inference systems to calculate the similarity forarrow The following description is the outline of the firstfuzzy inference system
e1
e2
e3
e4
(a)
e1
e2
e3
(b)
12059311205932
1205933 1205934
e1
e2
e3
e4
(c)
1205931
1205932
1205933
e1
e2
e3
(d)
Figure 7 Elements for arrows
(1) We first choose a cluster 119864 which includes fourstraight line elements denoted by 119890
1 1198902 1198903 and 119890
4
The four angles 1205931 1205932 1205933 and 120593
4are then measured
(see Figure 7(c))
(2) Next the following three attribute values are calcu-lated
(1) the standard deviation of 1205931 1205932 1205933 and 120593
4 it is
denoted by 1199091
(2) the average value of |1205931minus 1205934| and |120593
2minus 1205933| it is
denoted by 1199092
(3) the difference between the two lengths of theelements 119890
2and 1198904 it is denoted by 119909
3
(3) The fuzzy inference system is applied to 1199091 1199092 and
1199093 The fuzzy if-then rules are denoted below and
the membership functions are shown in Figure 8 Wethen have a value 119904 of [0 1] from the system 119904 impliesthe similarity for 119864
Rule 1 If 1199091is large 119909
2is small and 119909
3is small then it is
plausible that 119864 is an arrow
Rule 2 If 1199091is small then it is implausible that 119864 is an arrow
Rule 3 If 1199092is large then it is implausible that 119864 is an arrow
Rule 4 If 1199093is large then it is implausible that 119864 is an arrow
The second fuzzy inference system is applied to a cluster119864 which includes three straight line elements This systemhas the following three input attributes
(1) the standard deviation for 1205931 1205932 and 120593
3(see
Figure 7(d)) it is denoted by 1199091
(2) the difference between 1205932and 120593
3 it is denoted by 119909
2
(3) the difference between the two lengths of the elements1198901and 1198903 it is denoted by 119909
3
6 Advances in Fuzzy Systems
y = 1205831large (x1)y = 1205831small (x1)
y
10
0 10 30x1
(a) Membership functions for 1199091
y = 1205832large (x2)y = 1205832small (x2)
y
10
0 10 15x2
(b) Membership functions for 1199092
y = 1205833large (x3)y = 1205833small (x3)
y
10
0 10 20x3
(c) Membership functions for 1199093
y = 120583plau (x)y = 120583implau (x)
y
10
02 05 08x
0
(d) Membership functions for consequence
Figure 8Membership functions 120583119894small and 120583119894
large (119894 = 1 2 3) aremembership functions for the fuzzy sets small and large in Rule 119895 (119895 = 1 2 3)120583plau and 120583implau are membership functions for the fuzzy sets plausible and implausible in the consequence
e1
e2
e3
e4
(a)
e1
e2
e3
e4
12059311205932
1205933 1205934
(b)
Figure 9 Elements for cross
The following rules are the fuzzy if-then rules The member-ship functions are omitted
Rule 1 If 1199091is large 119909
2is small and 119909
3is small then it is
plausible that 119864 is an arrow
Rule 2 If 1199091is small then it is implausible that 119864 is an arrow
Rule 3 If 1199092is large then it is implausible that 119864 is an arrow
Rule 4 If 1199093is large then it is implausible that 119864 is an arrow
32 Cross Classification A cross consists of four straight lineelements (see Figure 9(a)) A cluster 119864 is applied to the fuzzyinference system for cross classification if 119864 includes four
straight line elements 1198901 1198902 1198903 and 119890
4 This fuzzy inference
system requires the following two input attributes(1) the standard deviation of 120593
1 1205932 1205933 and 120593
4(see
Figure 9(b)) it is denoted by 1199091
(2) the standard deviation for 1205931+ 1205932 1205932+ 1205933 1205933+ 1205934
and 1205931+ 1205934 it is denoted by 119909
2
The fuzzy if-then rules are given belowwhile themembershipfunctions are omitted
Rule 1 If 1199091is small and 119909
2is small then it is plausible that 119864
is a cross
Rule 2 If 1199091is large then it is implausible that 119864 is a cross
Rule 3 If 1199092is large then it is implausible that 119864 is a cross
If the similarity for a cluster obtained by the crossclassification is larger than a threshold value the cluster isclassified as a cross Similarly if the similarity from the arrowclassification exceeds a threshold value the cluster is classifiedas an arrow
33 Route and Railway Classification Route and railwayclassifications are conducted by the following way
(1) We first detect a sequence denoted by 119876 of clusterssuch that every two adjacent clusters include a com-mon straight line element (see Figure 10)
Advances in Fuzzy Systems 7
The principal line Arrowheads
e1 e2
(a) Route with three arrows the elements 1198901 and 1198902are common to the left and the right arrows
The principal line
e1 e2
Auxiliary lines
(b) Railway with three crosses the elements 1198901 and 1198902are common to the left and the right crosses
Figure 10 Route and railway symbols
Pi+1(xi+1 yi+1)
Pi(xi yi)
eiminus1
ei
120593iminus1
120593i
Piminus1(ximinus1 yiminus1)
Figure 11 Curvature for digital curve
(2) If the sequence 119876 consists of arrows then the routeclassification is executed to calculate the similaritydenoted by 119904
1 for the sequence119876 If 119904
1is larger than a
threshold value the sequence119876 is classified as a route(3) If the sequence119876 consists of crosses then the railway
classification is executed and we obtain the similaritydenoted by 119904
2 for the sequence 119876 If 119904
2exceeds a
threshold value the sequence 119876 is classified as arailway
Fuzzy inference systems are applied to compute thetwo similarities To calculate the attribute values for thefuzzy inference systems we detect the principal line andthe arrowhead (or the auxiliary lines) for the sequence 119876 asfollows If two straight line elements 119890
1and 119890
2 satisfy the
following two geometric characteristics it is plausible that 1198901
and 1198902are part of the same straight line
(1) The elements 1198901and 1198902are connected at an intersec-
tion 119875(2) The curvature at the intersection is small
Let119862 be a digital curve and let119875119894be a point on119862 A curvature
120581 of 119862 at 119875119894is then defined as the subtraction |120593
119894minus 120593119894minus1| (see
Figure 11) that is
120581 =1003816100381610038161003816120593119894 minus 120593119894minus1
1003816100381610038161003816 (4)
where 120593119894is the measure of an angle which is formed between
the 119909-axis and the line segment from 119875119894to 119875119894+1
and 120593119894minus1
isalso the measure of an angle which is formed between the 119909-axis and the line segment from 119875
119894minus1to 119875119894[24] Two adjacent
elements 119890119894minus1
and 119890119894are merged into a single straight line
segment if the curvature between 119890119894minus1
and 119890119894is less than a
threshold value After that we extract the principal line andthe arrowheads (or the auxiliary lines) for the sequence 119876
After extracting the principle line and the arrowhead (orthe auxiliary lines) for the sequence 119876 two fuzzy inferencesystems are applied to classify the sequence 119876 as a routeor railway The first fuzzy inference system is for routeclassification It has the following two input attributes
(1) the standard deviation for the lengths of elements inthe principal line it is denoted by 119909
1
(2) the standard deviation for lengths of elements in thearrowheads it is denoted by 119909
2
The fuzzy if-then rules for the first fuzzy inferencesystem are denoted below but the membership functions areomitted
Rule 1 If 1199091is small and 119909
2is small then it is plausible that119876
is a route
Rule 2 If 1199091is large then it is implausible that 119876 is a route
Rule 3 If 1199092is large then it is implausible that 119876 is a route
The second fuzzy inference system is for railway classifi-cation and the description for the second system is omittedbecause it is similar to the first system
4 SVG and Edel Documents Production
In our method a figure consists of elements whose shapesare as follows straight line circle circular arc and curveAs described in Section 22 the shape for an element isdetermined by the method of least squares If an element hasbeen classified as a curve then this element is expressed bya piecewise cubic Bezier curve the piecewise cubic Beziercurve is detected by an algorithmbased on themethodof leastsquares [25]
A Bezier curve 119876(119905) of degree 119899 is defined as
119876 (119905) =
119899
sum
119894=0
119881119894119861119899
119894(119905) (119905 isin [0 1]) (5)
where the 119881119894s are the control points and the 119861119899
119894s are the
Bernstein polynomials
119861119899
119894(119905) = (
119899
119894) 119905119894(1 minus 119905)
119899minus119894(119894 = 0 1 119899) (6)
8 Advances in Fuzzy Systems
Figure 12 A single cubic Bezier segment
0
2
4
6
8
0h lt 2h 2ndash5h 5ndash10h 10hlt
Figure 13 Experience in computers the vertical axis indicates thenumber of participants
Equation (5) is called a cubic Bezier curve when 119899 = 3Figure 12 shows an example of cubic Bezier curves the fourpoints119881
011988111198812 and119881
3are the control points the thick curve
is the cubic Bezier curve and the polygon expresses the oneconstructed by the four control points
SVG has the formats to express the four shapes andtherefore we can create an SVG document for a hand-drawnmap directly once we have detected the shapes for all theelements of the map On the other hand an Edel documentis a collection of embossed dots So it is easy to create anEdel document once we have detected the shapes for all theelements of a hand-drawn map
5 Experimental Results
This section describes the accuracy for our classificationsystem Five participants drew 15 maps using pencils andpapers These maps were then captured by using an imagescanner the resolution of the scannerwas set to 100 dpiThesedigital images were saved as 24-bits bitmap images The sizesof the images are in 1 169times850 pixelsWe havemeasured theaccuracy for our classification system using precision recalland f -measure
In the 15 maps there are 50 traffic signals 15 destinations15 departure places and 40 landmarks The classificationaccuracy for our system is summarized in Table 1 We canconclude that our system can produce the Edel and SVGdocuments almost correctly from hand-drawn maps
0
2
4
6
8
Freq
uent
ly
Som
etim
es
Occ
asio
nally
Seld
om
Nev
er
Figure 14 Experience in PPT the vertical axis indicates the numberof participants
Table 1 Experimental results are as follows 119873 is the number of allobjects for the symbol119879 is the number of objects correctly classifiedas the symbol 119862 is the number of objects classified as the symbol 119875is precision 119877 is recall and 119865 is 119865-measure
Symbol 119873 119879 119862 119875 () 119877 () 119865 ()Traffic signal 50 50 52 962 100 981Destination 15 15 15 100 100 100Departure place 15 14 14 100 933 965Landmark 40 38 38 100 950 974Cross 130 125 125 100 962 981Arrow 122 121 121 100 992 996Route 15 14 16 875 933 903Railway 28 28 28 100 100 100
6 Usability Evaluation for Our System
61 Method We have studied usability evaluation for oursystemThirteen participants 12 males and 1 female aged 20participated in this investigation all of them are the third-year university students To evaluate the usability for oursystem we selected two common methods for productionof tactile graphics one is the method to use the softwaresystem Edel which assists us to draw diagram images fortactile graphics produced by Braille embossers and anotherone is the method using swell papers In the second methoda diagram image is transferred to a swell paper by a printerand so forth A swell paper has been coated with thermallyfoamed microcapsules that respond to irradiation and causethe dark image on the paper to swell creating the tactilegraphic Microsoft PowerPoint (PPT) was selected to draw adiagram image on a swell paper because this software systemis commonly used in the universities in Japan
Before starting the experiment we asked the followingquestions Q1 Q2 and Q3 to all the participants and theresults for the questionsQ1 andQ3 are summarized in Figures13 and 14 We omit to show the result for the question Q2because all the participants have no experience in using EdelAll the participants are familiar with using computers andhave much experience in using PTT
Advances in Fuzzy Systems 9
(a) (b)
(c) (d)
(e) (f)
Figure 15 Maps for Sessions 1 and 2 (a) Edel map for Session 1 (b) PPT map for Session 1 (c) hand-drawn map for Session 1 (d) Edel mapfor Session 2 (e) PPT map for Session 2 and (f) hand-drawn map for Session 2
Q1 How long do you use computers every weekQ2 Have you ever created figures by using EdelQ3 Have you ever created figures by using PPT
We had two sessions in this investigation For the firstsession a participant created the map diagram shown inFigures 15(a)ndash15(c) the digital files were produced by usingEdel PPT and our system Note that in our system a userfirst draws a map manually using a pencil and paper thenconverts the hand-drawn map to a bitmap image using animage scanner and finally the bitmap image is transformed
into the two digital files the Edel and SVG documents Afterconducting all tasks the participant was asked to answer thefollowing two questions Q4 and Q5
Q4 Were you able to easily create the digital file(s) for thetactile map
Q5 Do you want to use this software system to create thedigital file for a tactile map
After the first session the participant had 5 minutes fortraining the software systems by himselfherself The secondsession followed this learning session In the second session
10 Advances in Fuzzy Systems
the participant conducted similar tasks to the first sessionbut the maps were those shown in Figures 15(d)ndash15(f) Aftercreating all digital files the participant was asked to answertwo questions Q4 and Q5 again and was also asked to answerthe following question Q6
Q6 Were you able to draw the map as you like
The orders of the software systems were determined ran-domly in both the first and the second sessions
62 Results We applied two-way ANOVA to the answersfor the 2 questions Q4 and Q5 The first factor denoted byFactor 1 is the difference in systems and the second factordenoted by Factor 2 is the difference in learning Factor 1includes three levels Edel PPT and our system and Factor2 includes two levels before learning and after learning Theresults from the two-way ANOVA for Q4 shows that therewas no significant interactions between Factor 1 and Factor2 (119865(2 72) = 092 119875 = 040) Furthermore there was asignificant difference for Factor 1 (119865(2 72) = 1926119875 lt 001)while no significant difference existed for Factor 2 (119865(1 72) =158 119875 = 021)The results for Q5 are similar to those for Q4That is there was no significant interaction between Factor1 and Factor 2 (119865(2 72) = 064 119875 = 053) there was asignificant difference for Factor 1 (119865(2 72) = 1400119875 lt 001)and therewas no significant difference for Factor 2 (119865(1 72) =132 119875 = 025)
We then conducted a multiple comparison test for theanswers of the questions Q4 and Q5 we selected Tukeyrsquoshonestly significant difference (HSD) criterion as themultiplecomparison test For Q4 there was a significant differencebetween Edel and our system (119875 lt 001) and there was alsoa significant difference between PPT and our system (119875 lt
001) however there was no significant difference betweenEdel and PPT (see Figure 16) Furthermore for Q5 therewas a significant difference between Edel and our system(119875 lt 001) and therewas also a significant difference betweenPPT and our system (119875 lt 005) however we observed amarginally significant difference between Edel and PPT (119875 lt01) (see Figure 17)
Lastly we applied a one-way ANOVA to the answersfor the question Q6 the factor of this one-way ANOVA isthe difference between systems The results are shown inFigure 18 and the results for the one-way ANOVA show thatthere was a marginally significant difference (119865(2 36) = 259119875 = 0089) We then applied Tukeyrsquos HSD and we observeda marginally significant difference between Edel and oursystem
As a result of the discussion above we can concludethat users feel that our method is easier to produce tactilemaps than the conventional twomethods Furthermore mapimages produced by our system are visually as fine as mapimages produced by Edel and PPT
7 Conclusions
In this paper a computer-aided system for automating pro-duction of tactile graphics from hand-drawn maps has been
0
2
4
6
Edel PPT Our System
Scor
e
Before LearningAfter Learning
lowastlowastlowastlowast
Figure 16 Results for Q4 the average and standard error for eachsystem (lowastlowast119875 lt 001)
0
1
2
3
4
Edel PPT Our System
Scor
e
lowastlowastlowast
Before LearningAfter Learning
Figure 17 Results for Q5 the average and standard error for eachsystem (lowastlowast119875 lt 001 lowast119875 lt 005)
developed The system includes the following four mainprocedures (1) preprocessing (2) segmentation (3) patternrecognition based on fuzzy inference and (4) SVG and Edeldocuments production The preprocessing is applied to aninitial hand-drawn map in order to remove noise and thena binary image is obtained In the segmentation procedurethe binary image is first skeletonized and then the skeletonis segmented into elements by eliminating intersections andcorner points Then a pattern recognition method is appliedto recognize symbols Lastly SVG and Edel documents arecreated to save the recognition results The usability for oursystem was evaluated by comparing our method with the twoconventional systems for creating tactile maps The resultsshow that our system is easier to create tactile maps than theother two systems
In our fuzzy inference systems triangular and trapezoidalmembership functions are applied to define the fuzzy setsdue to their simplicity The shapes of membership functionsinfluence the accuracy of recognition results How to choosean optimal membership function is our future work In thisstudy we do not assume a map includes character stringsHowever adding character strings to a map is very important
Advances in Fuzzy Systems 11
0
1
2
3
4
Edel PPT Our System
Scor
e
Figure 18 Results for Q6 the average and standard error for eachsystem
because it increases the comprehension for the map In thepresent work the types of symbols used in hand-drawnmaps are restricted Developing a system for automatingtranslation of hand-drawn maps with character string andvarious symbols is one of our future works
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This work was supported by the JSPS Grant-in-Aid forScientific Research (C) no 24501198 The authors thank DrKeisuke Ido for giving much advice to the experiments ofusability evaluation
References
[1] J A Miele S Landau and D Gilden ldquoTalking TMAPautomated generation of audio-tactile maps using Smith-Kettlewellrsquos TMAP softwarerdquo The British Journal of VisualImpairment vol 24 no 2 pp 93ndash100 2006
[2] T Watanabe T Yamaguchi K Watanabe et al ldquoDevelopmentand evaluation of a tactile map automated creation systemaccessible to blind personsrdquo IEICE Transactions on Informationand Systems vol J94-D no 10 pp 1652ndash1663 2011 (Japanese)
[3] TMACS httptmacsengniigata-uacjptmacs-dev[4] httptenpuchizugsigojpshokuchizu[5] S E Krufka and K E Barner ldquoAutomatic production of tactile
graphics from scalable vector graphicsrdquo in Proceedings of the 7thInternational ACM SIGACCESS Conference on Computers andAccessibility (ASSETS rsquo05) pp 166ndash172 October 2005
[6] R E Ladner M Y Ivory R Rao et al ldquoAutomating tactilegraphics translationrdquo in Proceedings of the 7th InternationalACMSIGACCESSConference onComputer andAccessibility pp150ndash157 New York NY USA October 2005
[7] Edel httpwww7abiglobenejp EDEL-plus[8] J D Eisenberg SVG Essentials OrsquoREILLY 2002[9] K Broelemann X Jiang and A Schwering ldquoAutomatic street
graph construction in sketchmapsrdquo inGraph-Based Representa-tions in Pattern Recognition X Jiang M Ferrer and A Torsello
Eds vol 6658 of Lecture Notes in Computer Science pp 275ndash284 Springer Berlin Germany 2011
[10] L B Kara and T F Stahovich ldquoAn image-based trainablesymbol recognizer for hand-drawn sketchesrdquo Computers ampGraphics vol 29 no 4 pp 501ndash517 2005
[11] B Edwards and V Chandran ldquoMachine recognition of hand-drawn circuit diagramsrdquo in Proceedings of the IEEE InterntionalConference onAcoustics Speech and Signal Processing vol 6 pp3618ndash3621 June 2000
[12] P Sala ldquoA recognition system for symbols of electronic com-ponents in hand-written circuit diagramsrdquo CSC 2515-MachineLearning Project Report 2004
[13] Y Qiao and M Yasuhara ldquoRecovering dynamic informationfrom static handwritten imagesrdquo in Proceedings of the 9th Inter-national Workshop on Frontiers in Handwriting Recognition pp118ndash123 jpn October 2004
[14] M Notowidigdo and R C Miller ldquoOff-line sketch interpreta-tionrdquo in Proceedings of the AAAI Fall Symposium on MakingPen-Based Interaction Intelligent and Natural pp 120ndash126Arlington Va USA April 2004
[15] R J Gonzalez and R E Woods Digital Image ProcessingPearson Education Upper Saddle River NJ USA 3rd edition2007
[16] N Otsu ldquoA threshold selection method from gray-level his-togramsrdquo IEEE Transactions on Systems Man and Cyberneticsvol 9 no 1 pp 62ndash66 1979
[17] C J Hilditch ldquoLinear skeletons from square cupboardsrdquo inMachine Intelligence Vol IV B Meltzer andDMichie Eds pp403ndash420 Elsevier New York NY USA 1969
[18] S Hermann and R Klette ldquoGlobal curvature estimation for cor-ner detectionrdquo Communication and Information TechnologyResearch Technical Report 171 2005
[19] N Nain V Laxmi B Bhadviya and A Gopal ldquoCorner detec-tion using difference chain code as curvaturerdquo in Proceedingsof the 3rd IEEE International Conference on Signal ImageTechnologies and Internet Based Systems (SITIS 07) pp 821ndash825Shanghai China December 2007
[20] B Kerautret J Lachaud and B Naegel ldquoCurvature basedcorner detector for discrete noisy and multi-scale contoursrdquoInternational Journal of Shape Modeling vol 14 no 2 pp 127ndash145 2008
[21] K Wall and P Danielsson ldquoA fast sequential method forpolygonal approximation of digitized curvesrdquoComputer VisionGraphics amp Image Processing vol 28 no 2 pp 220ndash227 1984
[22] D Chetverikov ldquoA simple and efficient algorithm for detectionof high curvature points in planar curvesrdquo inComputer Analysisof Images and Patterns vol 2756 of Lecture Notes in ComputerScience pp 746ndash753 Springer Berlin Germany 2003
[23] T Terano K Asai and M Sugeno Fuzzy Systems and ItsApplications Academic Press Boston Mass USA 2nd edition1992
[24] N Ono and R Takiyama ldquoOn calculations of curvature ofsampled curvesrdquo Technical Report of IEICE IE93-74 1993(Japanese)
[25] P J Schneider ldquoAn algorithm for automatically fitting digitizedcurvesrdquo in Graphics Gems A S Glassner Ed pp 612ndash625Academic Press 1990
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
6 Advances in Fuzzy Systems
y = 1205831large (x1)y = 1205831small (x1)
y
10
0 10 30x1
(a) Membership functions for 1199091
y = 1205832large (x2)y = 1205832small (x2)
y
10
0 10 15x2
(b) Membership functions for 1199092
y = 1205833large (x3)y = 1205833small (x3)
y
10
0 10 20x3
(c) Membership functions for 1199093
y = 120583plau (x)y = 120583implau (x)
y
10
02 05 08x
0
(d) Membership functions for consequence
Figure 8Membership functions 120583119894small and 120583119894
large (119894 = 1 2 3) aremembership functions for the fuzzy sets small and large in Rule 119895 (119895 = 1 2 3)120583plau and 120583implau are membership functions for the fuzzy sets plausible and implausible in the consequence
e1
e2
e3
e4
(a)
e1
e2
e3
e4
12059311205932
1205933 1205934
(b)
Figure 9 Elements for cross
The following rules are the fuzzy if-then rules The member-ship functions are omitted
Rule 1 If 1199091is large 119909
2is small and 119909
3is small then it is
plausible that 119864 is an arrow
Rule 2 If 1199091is small then it is implausible that 119864 is an arrow
Rule 3 If 1199092is large then it is implausible that 119864 is an arrow
Rule 4 If 1199093is large then it is implausible that 119864 is an arrow
32 Cross Classification A cross consists of four straight lineelements (see Figure 9(a)) A cluster 119864 is applied to the fuzzyinference system for cross classification if 119864 includes four
straight line elements 1198901 1198902 1198903 and 119890
4 This fuzzy inference
system requires the following two input attributes(1) the standard deviation of 120593
1 1205932 1205933 and 120593
4(see
Figure 9(b)) it is denoted by 1199091
(2) the standard deviation for 1205931+ 1205932 1205932+ 1205933 1205933+ 1205934
and 1205931+ 1205934 it is denoted by 119909
2
The fuzzy if-then rules are given belowwhile themembershipfunctions are omitted
Rule 1 If 1199091is small and 119909
2is small then it is plausible that 119864
is a cross
Rule 2 If 1199091is large then it is implausible that 119864 is a cross
Rule 3 If 1199092is large then it is implausible that 119864 is a cross
If the similarity for a cluster obtained by the crossclassification is larger than a threshold value the cluster isclassified as a cross Similarly if the similarity from the arrowclassification exceeds a threshold value the cluster is classifiedas an arrow
33 Route and Railway Classification Route and railwayclassifications are conducted by the following way
(1) We first detect a sequence denoted by 119876 of clusterssuch that every two adjacent clusters include a com-mon straight line element (see Figure 10)
Advances in Fuzzy Systems 7
The principal line Arrowheads
e1 e2
(a) Route with three arrows the elements 1198901 and 1198902are common to the left and the right arrows
The principal line
e1 e2
Auxiliary lines
(b) Railway with three crosses the elements 1198901 and 1198902are common to the left and the right crosses
Figure 10 Route and railway symbols
Pi+1(xi+1 yi+1)
Pi(xi yi)
eiminus1
ei
120593iminus1
120593i
Piminus1(ximinus1 yiminus1)
Figure 11 Curvature for digital curve
(2) If the sequence 119876 consists of arrows then the routeclassification is executed to calculate the similaritydenoted by 119904
1 for the sequence119876 If 119904
1is larger than a
threshold value the sequence119876 is classified as a route(3) If the sequence119876 consists of crosses then the railway
classification is executed and we obtain the similaritydenoted by 119904
2 for the sequence 119876 If 119904
2exceeds a
threshold value the sequence 119876 is classified as arailway
Fuzzy inference systems are applied to compute thetwo similarities To calculate the attribute values for thefuzzy inference systems we detect the principal line andthe arrowhead (or the auxiliary lines) for the sequence 119876 asfollows If two straight line elements 119890
1and 119890
2 satisfy the
following two geometric characteristics it is plausible that 1198901
and 1198902are part of the same straight line
(1) The elements 1198901and 1198902are connected at an intersec-
tion 119875(2) The curvature at the intersection is small
Let119862 be a digital curve and let119875119894be a point on119862 A curvature
120581 of 119862 at 119875119894is then defined as the subtraction |120593
119894minus 120593119894minus1| (see
Figure 11) that is
120581 =1003816100381610038161003816120593119894 minus 120593119894minus1
1003816100381610038161003816 (4)
where 120593119894is the measure of an angle which is formed between
the 119909-axis and the line segment from 119875119894to 119875119894+1
and 120593119894minus1
isalso the measure of an angle which is formed between the 119909-axis and the line segment from 119875
119894minus1to 119875119894[24] Two adjacent
elements 119890119894minus1
and 119890119894are merged into a single straight line
segment if the curvature between 119890119894minus1
and 119890119894is less than a
threshold value After that we extract the principal line andthe arrowheads (or the auxiliary lines) for the sequence 119876
After extracting the principle line and the arrowhead (orthe auxiliary lines) for the sequence 119876 two fuzzy inferencesystems are applied to classify the sequence 119876 as a routeor railway The first fuzzy inference system is for routeclassification It has the following two input attributes
(1) the standard deviation for the lengths of elements inthe principal line it is denoted by 119909
1
(2) the standard deviation for lengths of elements in thearrowheads it is denoted by 119909
2
The fuzzy if-then rules for the first fuzzy inferencesystem are denoted below but the membership functions areomitted
Rule 1 If 1199091is small and 119909
2is small then it is plausible that119876
is a route
Rule 2 If 1199091is large then it is implausible that 119876 is a route
Rule 3 If 1199092is large then it is implausible that 119876 is a route
The second fuzzy inference system is for railway classifi-cation and the description for the second system is omittedbecause it is similar to the first system
4 SVG and Edel Documents Production
In our method a figure consists of elements whose shapesare as follows straight line circle circular arc and curveAs described in Section 22 the shape for an element isdetermined by the method of least squares If an element hasbeen classified as a curve then this element is expressed bya piecewise cubic Bezier curve the piecewise cubic Beziercurve is detected by an algorithmbased on themethodof leastsquares [25]
A Bezier curve 119876(119905) of degree 119899 is defined as
119876 (119905) =
119899
sum
119894=0
119881119894119861119899
119894(119905) (119905 isin [0 1]) (5)
where the 119881119894s are the control points and the 119861119899
119894s are the
Bernstein polynomials
119861119899
119894(119905) = (
119899
119894) 119905119894(1 minus 119905)
119899minus119894(119894 = 0 1 119899) (6)
8 Advances in Fuzzy Systems
Figure 12 A single cubic Bezier segment
0
2
4
6
8
0h lt 2h 2ndash5h 5ndash10h 10hlt
Figure 13 Experience in computers the vertical axis indicates thenumber of participants
Equation (5) is called a cubic Bezier curve when 119899 = 3Figure 12 shows an example of cubic Bezier curves the fourpoints119881
011988111198812 and119881
3are the control points the thick curve
is the cubic Bezier curve and the polygon expresses the oneconstructed by the four control points
SVG has the formats to express the four shapes andtherefore we can create an SVG document for a hand-drawnmap directly once we have detected the shapes for all theelements of the map On the other hand an Edel documentis a collection of embossed dots So it is easy to create anEdel document once we have detected the shapes for all theelements of a hand-drawn map
5 Experimental Results
This section describes the accuracy for our classificationsystem Five participants drew 15 maps using pencils andpapers These maps were then captured by using an imagescanner the resolution of the scannerwas set to 100 dpiThesedigital images were saved as 24-bits bitmap images The sizesof the images are in 1 169times850 pixelsWe havemeasured theaccuracy for our classification system using precision recalland f -measure
In the 15 maps there are 50 traffic signals 15 destinations15 departure places and 40 landmarks The classificationaccuracy for our system is summarized in Table 1 We canconclude that our system can produce the Edel and SVGdocuments almost correctly from hand-drawn maps
0
2
4
6
8
Freq
uent
ly
Som
etim
es
Occ
asio
nally
Seld
om
Nev
er
Figure 14 Experience in PPT the vertical axis indicates the numberof participants
Table 1 Experimental results are as follows 119873 is the number of allobjects for the symbol119879 is the number of objects correctly classifiedas the symbol 119862 is the number of objects classified as the symbol 119875is precision 119877 is recall and 119865 is 119865-measure
Symbol 119873 119879 119862 119875 () 119877 () 119865 ()Traffic signal 50 50 52 962 100 981Destination 15 15 15 100 100 100Departure place 15 14 14 100 933 965Landmark 40 38 38 100 950 974Cross 130 125 125 100 962 981Arrow 122 121 121 100 992 996Route 15 14 16 875 933 903Railway 28 28 28 100 100 100
6 Usability Evaluation for Our System
61 Method We have studied usability evaluation for oursystemThirteen participants 12 males and 1 female aged 20participated in this investigation all of them are the third-year university students To evaluate the usability for oursystem we selected two common methods for productionof tactile graphics one is the method to use the softwaresystem Edel which assists us to draw diagram images fortactile graphics produced by Braille embossers and anotherone is the method using swell papers In the second methoda diagram image is transferred to a swell paper by a printerand so forth A swell paper has been coated with thermallyfoamed microcapsules that respond to irradiation and causethe dark image on the paper to swell creating the tactilegraphic Microsoft PowerPoint (PPT) was selected to draw adiagram image on a swell paper because this software systemis commonly used in the universities in Japan
Before starting the experiment we asked the followingquestions Q1 Q2 and Q3 to all the participants and theresults for the questionsQ1 andQ3 are summarized in Figures13 and 14 We omit to show the result for the question Q2because all the participants have no experience in using EdelAll the participants are familiar with using computers andhave much experience in using PTT
Advances in Fuzzy Systems 9
(a) (b)
(c) (d)
(e) (f)
Figure 15 Maps for Sessions 1 and 2 (a) Edel map for Session 1 (b) PPT map for Session 1 (c) hand-drawn map for Session 1 (d) Edel mapfor Session 2 (e) PPT map for Session 2 and (f) hand-drawn map for Session 2
Q1 How long do you use computers every weekQ2 Have you ever created figures by using EdelQ3 Have you ever created figures by using PPT
We had two sessions in this investigation For the firstsession a participant created the map diagram shown inFigures 15(a)ndash15(c) the digital files were produced by usingEdel PPT and our system Note that in our system a userfirst draws a map manually using a pencil and paper thenconverts the hand-drawn map to a bitmap image using animage scanner and finally the bitmap image is transformed
into the two digital files the Edel and SVG documents Afterconducting all tasks the participant was asked to answer thefollowing two questions Q4 and Q5
Q4 Were you able to easily create the digital file(s) for thetactile map
Q5 Do you want to use this software system to create thedigital file for a tactile map
After the first session the participant had 5 minutes fortraining the software systems by himselfherself The secondsession followed this learning session In the second session
10 Advances in Fuzzy Systems
the participant conducted similar tasks to the first sessionbut the maps were those shown in Figures 15(d)ndash15(f) Aftercreating all digital files the participant was asked to answertwo questions Q4 and Q5 again and was also asked to answerthe following question Q6
Q6 Were you able to draw the map as you like
The orders of the software systems were determined ran-domly in both the first and the second sessions
62 Results We applied two-way ANOVA to the answersfor the 2 questions Q4 and Q5 The first factor denoted byFactor 1 is the difference in systems and the second factordenoted by Factor 2 is the difference in learning Factor 1includes three levels Edel PPT and our system and Factor2 includes two levels before learning and after learning Theresults from the two-way ANOVA for Q4 shows that therewas no significant interactions between Factor 1 and Factor2 (119865(2 72) = 092 119875 = 040) Furthermore there was asignificant difference for Factor 1 (119865(2 72) = 1926119875 lt 001)while no significant difference existed for Factor 2 (119865(1 72) =158 119875 = 021)The results for Q5 are similar to those for Q4That is there was no significant interaction between Factor1 and Factor 2 (119865(2 72) = 064 119875 = 053) there was asignificant difference for Factor 1 (119865(2 72) = 1400119875 lt 001)and therewas no significant difference for Factor 2 (119865(1 72) =132 119875 = 025)
We then conducted a multiple comparison test for theanswers of the questions Q4 and Q5 we selected Tukeyrsquoshonestly significant difference (HSD) criterion as themultiplecomparison test For Q4 there was a significant differencebetween Edel and our system (119875 lt 001) and there was alsoa significant difference between PPT and our system (119875 lt
001) however there was no significant difference betweenEdel and PPT (see Figure 16) Furthermore for Q5 therewas a significant difference between Edel and our system(119875 lt 001) and therewas also a significant difference betweenPPT and our system (119875 lt 005) however we observed amarginally significant difference between Edel and PPT (119875 lt01) (see Figure 17)
Lastly we applied a one-way ANOVA to the answersfor the question Q6 the factor of this one-way ANOVA isthe difference between systems The results are shown inFigure 18 and the results for the one-way ANOVA show thatthere was a marginally significant difference (119865(2 36) = 259119875 = 0089) We then applied Tukeyrsquos HSD and we observeda marginally significant difference between Edel and oursystem
As a result of the discussion above we can concludethat users feel that our method is easier to produce tactilemaps than the conventional twomethods Furthermore mapimages produced by our system are visually as fine as mapimages produced by Edel and PPT
7 Conclusions
In this paper a computer-aided system for automating pro-duction of tactile graphics from hand-drawn maps has been
0
2
4
6
Edel PPT Our System
Scor
e
Before LearningAfter Learning
lowastlowastlowastlowast
Figure 16 Results for Q4 the average and standard error for eachsystem (lowastlowast119875 lt 001)
0
1
2
3
4
Edel PPT Our System
Scor
e
lowastlowastlowast
Before LearningAfter Learning
Figure 17 Results for Q5 the average and standard error for eachsystem (lowastlowast119875 lt 001 lowast119875 lt 005)
developed The system includes the following four mainprocedures (1) preprocessing (2) segmentation (3) patternrecognition based on fuzzy inference and (4) SVG and Edeldocuments production The preprocessing is applied to aninitial hand-drawn map in order to remove noise and thena binary image is obtained In the segmentation procedurethe binary image is first skeletonized and then the skeletonis segmented into elements by eliminating intersections andcorner points Then a pattern recognition method is appliedto recognize symbols Lastly SVG and Edel documents arecreated to save the recognition results The usability for oursystem was evaluated by comparing our method with the twoconventional systems for creating tactile maps The resultsshow that our system is easier to create tactile maps than theother two systems
In our fuzzy inference systems triangular and trapezoidalmembership functions are applied to define the fuzzy setsdue to their simplicity The shapes of membership functionsinfluence the accuracy of recognition results How to choosean optimal membership function is our future work In thisstudy we do not assume a map includes character stringsHowever adding character strings to a map is very important
Advances in Fuzzy Systems 11
0
1
2
3
4
Edel PPT Our System
Scor
e
Figure 18 Results for Q6 the average and standard error for eachsystem
because it increases the comprehension for the map In thepresent work the types of symbols used in hand-drawnmaps are restricted Developing a system for automatingtranslation of hand-drawn maps with character string andvarious symbols is one of our future works
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This work was supported by the JSPS Grant-in-Aid forScientific Research (C) no 24501198 The authors thank DrKeisuke Ido for giving much advice to the experiments ofusability evaluation
References
[1] J A Miele S Landau and D Gilden ldquoTalking TMAPautomated generation of audio-tactile maps using Smith-Kettlewellrsquos TMAP softwarerdquo The British Journal of VisualImpairment vol 24 no 2 pp 93ndash100 2006
[2] T Watanabe T Yamaguchi K Watanabe et al ldquoDevelopmentand evaluation of a tactile map automated creation systemaccessible to blind personsrdquo IEICE Transactions on Informationand Systems vol J94-D no 10 pp 1652ndash1663 2011 (Japanese)
[3] TMACS httptmacsengniigata-uacjptmacs-dev[4] httptenpuchizugsigojpshokuchizu[5] S E Krufka and K E Barner ldquoAutomatic production of tactile
graphics from scalable vector graphicsrdquo in Proceedings of the 7thInternational ACM SIGACCESS Conference on Computers andAccessibility (ASSETS rsquo05) pp 166ndash172 October 2005
[6] R E Ladner M Y Ivory R Rao et al ldquoAutomating tactilegraphics translationrdquo in Proceedings of the 7th InternationalACMSIGACCESSConference onComputer andAccessibility pp150ndash157 New York NY USA October 2005
[7] Edel httpwww7abiglobenejp EDEL-plus[8] J D Eisenberg SVG Essentials OrsquoREILLY 2002[9] K Broelemann X Jiang and A Schwering ldquoAutomatic street
graph construction in sketchmapsrdquo inGraph-Based Representa-tions in Pattern Recognition X Jiang M Ferrer and A Torsello
Eds vol 6658 of Lecture Notes in Computer Science pp 275ndash284 Springer Berlin Germany 2011
[10] L B Kara and T F Stahovich ldquoAn image-based trainablesymbol recognizer for hand-drawn sketchesrdquo Computers ampGraphics vol 29 no 4 pp 501ndash517 2005
[11] B Edwards and V Chandran ldquoMachine recognition of hand-drawn circuit diagramsrdquo in Proceedings of the IEEE InterntionalConference onAcoustics Speech and Signal Processing vol 6 pp3618ndash3621 June 2000
[12] P Sala ldquoA recognition system for symbols of electronic com-ponents in hand-written circuit diagramsrdquo CSC 2515-MachineLearning Project Report 2004
[13] Y Qiao and M Yasuhara ldquoRecovering dynamic informationfrom static handwritten imagesrdquo in Proceedings of the 9th Inter-national Workshop on Frontiers in Handwriting Recognition pp118ndash123 jpn October 2004
[14] M Notowidigdo and R C Miller ldquoOff-line sketch interpreta-tionrdquo in Proceedings of the AAAI Fall Symposium on MakingPen-Based Interaction Intelligent and Natural pp 120ndash126Arlington Va USA April 2004
[15] R J Gonzalez and R E Woods Digital Image ProcessingPearson Education Upper Saddle River NJ USA 3rd edition2007
[16] N Otsu ldquoA threshold selection method from gray-level his-togramsrdquo IEEE Transactions on Systems Man and Cyberneticsvol 9 no 1 pp 62ndash66 1979
[17] C J Hilditch ldquoLinear skeletons from square cupboardsrdquo inMachine Intelligence Vol IV B Meltzer andDMichie Eds pp403ndash420 Elsevier New York NY USA 1969
[18] S Hermann and R Klette ldquoGlobal curvature estimation for cor-ner detectionrdquo Communication and Information TechnologyResearch Technical Report 171 2005
[19] N Nain V Laxmi B Bhadviya and A Gopal ldquoCorner detec-tion using difference chain code as curvaturerdquo in Proceedingsof the 3rd IEEE International Conference on Signal ImageTechnologies and Internet Based Systems (SITIS 07) pp 821ndash825Shanghai China December 2007
[20] B Kerautret J Lachaud and B Naegel ldquoCurvature basedcorner detector for discrete noisy and multi-scale contoursrdquoInternational Journal of Shape Modeling vol 14 no 2 pp 127ndash145 2008
[21] K Wall and P Danielsson ldquoA fast sequential method forpolygonal approximation of digitized curvesrdquoComputer VisionGraphics amp Image Processing vol 28 no 2 pp 220ndash227 1984
[22] D Chetverikov ldquoA simple and efficient algorithm for detectionof high curvature points in planar curvesrdquo inComputer Analysisof Images and Patterns vol 2756 of Lecture Notes in ComputerScience pp 746ndash753 Springer Berlin Germany 2003
[23] T Terano K Asai and M Sugeno Fuzzy Systems and ItsApplications Academic Press Boston Mass USA 2nd edition1992
[24] N Ono and R Takiyama ldquoOn calculations of curvature ofsampled curvesrdquo Technical Report of IEICE IE93-74 1993(Japanese)
[25] P J Schneider ldquoAn algorithm for automatically fitting digitizedcurvesrdquo in Graphics Gems A S Glassner Ed pp 612ndash625Academic Press 1990
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Advances in Fuzzy Systems 7
The principal line Arrowheads
e1 e2
(a) Route with three arrows the elements 1198901 and 1198902are common to the left and the right arrows
The principal line
e1 e2
Auxiliary lines
(b) Railway with three crosses the elements 1198901 and 1198902are common to the left and the right crosses
Figure 10 Route and railway symbols
Pi+1(xi+1 yi+1)
Pi(xi yi)
eiminus1
ei
120593iminus1
120593i
Piminus1(ximinus1 yiminus1)
Figure 11 Curvature for digital curve
(2) If the sequence 119876 consists of arrows then the routeclassification is executed to calculate the similaritydenoted by 119904
1 for the sequence119876 If 119904
1is larger than a
threshold value the sequence119876 is classified as a route(3) If the sequence119876 consists of crosses then the railway
classification is executed and we obtain the similaritydenoted by 119904
2 for the sequence 119876 If 119904
2exceeds a
threshold value the sequence 119876 is classified as arailway
Fuzzy inference systems are applied to compute thetwo similarities To calculate the attribute values for thefuzzy inference systems we detect the principal line andthe arrowhead (or the auxiliary lines) for the sequence 119876 asfollows If two straight line elements 119890
1and 119890
2 satisfy the
following two geometric characteristics it is plausible that 1198901
and 1198902are part of the same straight line
(1) The elements 1198901and 1198902are connected at an intersec-
tion 119875(2) The curvature at the intersection is small
Let119862 be a digital curve and let119875119894be a point on119862 A curvature
120581 of 119862 at 119875119894is then defined as the subtraction |120593
119894minus 120593119894minus1| (see
Figure 11) that is
120581 =1003816100381610038161003816120593119894 minus 120593119894minus1
1003816100381610038161003816 (4)
where 120593119894is the measure of an angle which is formed between
the 119909-axis and the line segment from 119875119894to 119875119894+1
and 120593119894minus1
isalso the measure of an angle which is formed between the 119909-axis and the line segment from 119875
119894minus1to 119875119894[24] Two adjacent
elements 119890119894minus1
and 119890119894are merged into a single straight line
segment if the curvature between 119890119894minus1
and 119890119894is less than a
threshold value After that we extract the principal line andthe arrowheads (or the auxiliary lines) for the sequence 119876
After extracting the principle line and the arrowhead (orthe auxiliary lines) for the sequence 119876 two fuzzy inferencesystems are applied to classify the sequence 119876 as a routeor railway The first fuzzy inference system is for routeclassification It has the following two input attributes
(1) the standard deviation for the lengths of elements inthe principal line it is denoted by 119909
1
(2) the standard deviation for lengths of elements in thearrowheads it is denoted by 119909
2
The fuzzy if-then rules for the first fuzzy inferencesystem are denoted below but the membership functions areomitted
Rule 1 If 1199091is small and 119909
2is small then it is plausible that119876
is a route
Rule 2 If 1199091is large then it is implausible that 119876 is a route
Rule 3 If 1199092is large then it is implausible that 119876 is a route
The second fuzzy inference system is for railway classifi-cation and the description for the second system is omittedbecause it is similar to the first system
4 SVG and Edel Documents Production
In our method a figure consists of elements whose shapesare as follows straight line circle circular arc and curveAs described in Section 22 the shape for an element isdetermined by the method of least squares If an element hasbeen classified as a curve then this element is expressed bya piecewise cubic Bezier curve the piecewise cubic Beziercurve is detected by an algorithmbased on themethodof leastsquares [25]
A Bezier curve 119876(119905) of degree 119899 is defined as
119876 (119905) =
119899
sum
119894=0
119881119894119861119899
119894(119905) (119905 isin [0 1]) (5)
where the 119881119894s are the control points and the 119861119899
119894s are the
Bernstein polynomials
119861119899
119894(119905) = (
119899
119894) 119905119894(1 minus 119905)
119899minus119894(119894 = 0 1 119899) (6)
8 Advances in Fuzzy Systems
Figure 12 A single cubic Bezier segment
0
2
4
6
8
0h lt 2h 2ndash5h 5ndash10h 10hlt
Figure 13 Experience in computers the vertical axis indicates thenumber of participants
Equation (5) is called a cubic Bezier curve when 119899 = 3Figure 12 shows an example of cubic Bezier curves the fourpoints119881
011988111198812 and119881
3are the control points the thick curve
is the cubic Bezier curve and the polygon expresses the oneconstructed by the four control points
SVG has the formats to express the four shapes andtherefore we can create an SVG document for a hand-drawnmap directly once we have detected the shapes for all theelements of the map On the other hand an Edel documentis a collection of embossed dots So it is easy to create anEdel document once we have detected the shapes for all theelements of a hand-drawn map
5 Experimental Results
This section describes the accuracy for our classificationsystem Five participants drew 15 maps using pencils andpapers These maps were then captured by using an imagescanner the resolution of the scannerwas set to 100 dpiThesedigital images were saved as 24-bits bitmap images The sizesof the images are in 1 169times850 pixelsWe havemeasured theaccuracy for our classification system using precision recalland f -measure
In the 15 maps there are 50 traffic signals 15 destinations15 departure places and 40 landmarks The classificationaccuracy for our system is summarized in Table 1 We canconclude that our system can produce the Edel and SVGdocuments almost correctly from hand-drawn maps
0
2
4
6
8
Freq
uent
ly
Som
etim
es
Occ
asio
nally
Seld
om
Nev
er
Figure 14 Experience in PPT the vertical axis indicates the numberof participants
Table 1 Experimental results are as follows 119873 is the number of allobjects for the symbol119879 is the number of objects correctly classifiedas the symbol 119862 is the number of objects classified as the symbol 119875is precision 119877 is recall and 119865 is 119865-measure
Symbol 119873 119879 119862 119875 () 119877 () 119865 ()Traffic signal 50 50 52 962 100 981Destination 15 15 15 100 100 100Departure place 15 14 14 100 933 965Landmark 40 38 38 100 950 974Cross 130 125 125 100 962 981Arrow 122 121 121 100 992 996Route 15 14 16 875 933 903Railway 28 28 28 100 100 100
6 Usability Evaluation for Our System
61 Method We have studied usability evaluation for oursystemThirteen participants 12 males and 1 female aged 20participated in this investigation all of them are the third-year university students To evaluate the usability for oursystem we selected two common methods for productionof tactile graphics one is the method to use the softwaresystem Edel which assists us to draw diagram images fortactile graphics produced by Braille embossers and anotherone is the method using swell papers In the second methoda diagram image is transferred to a swell paper by a printerand so forth A swell paper has been coated with thermallyfoamed microcapsules that respond to irradiation and causethe dark image on the paper to swell creating the tactilegraphic Microsoft PowerPoint (PPT) was selected to draw adiagram image on a swell paper because this software systemis commonly used in the universities in Japan
Before starting the experiment we asked the followingquestions Q1 Q2 and Q3 to all the participants and theresults for the questionsQ1 andQ3 are summarized in Figures13 and 14 We omit to show the result for the question Q2because all the participants have no experience in using EdelAll the participants are familiar with using computers andhave much experience in using PTT
Advances in Fuzzy Systems 9
(a) (b)
(c) (d)
(e) (f)
Figure 15 Maps for Sessions 1 and 2 (a) Edel map for Session 1 (b) PPT map for Session 1 (c) hand-drawn map for Session 1 (d) Edel mapfor Session 2 (e) PPT map for Session 2 and (f) hand-drawn map for Session 2
Q1 How long do you use computers every weekQ2 Have you ever created figures by using EdelQ3 Have you ever created figures by using PPT
We had two sessions in this investigation For the firstsession a participant created the map diagram shown inFigures 15(a)ndash15(c) the digital files were produced by usingEdel PPT and our system Note that in our system a userfirst draws a map manually using a pencil and paper thenconverts the hand-drawn map to a bitmap image using animage scanner and finally the bitmap image is transformed
into the two digital files the Edel and SVG documents Afterconducting all tasks the participant was asked to answer thefollowing two questions Q4 and Q5
Q4 Were you able to easily create the digital file(s) for thetactile map
Q5 Do you want to use this software system to create thedigital file for a tactile map
After the first session the participant had 5 minutes fortraining the software systems by himselfherself The secondsession followed this learning session In the second session
10 Advances in Fuzzy Systems
the participant conducted similar tasks to the first sessionbut the maps were those shown in Figures 15(d)ndash15(f) Aftercreating all digital files the participant was asked to answertwo questions Q4 and Q5 again and was also asked to answerthe following question Q6
Q6 Were you able to draw the map as you like
The orders of the software systems were determined ran-domly in both the first and the second sessions
62 Results We applied two-way ANOVA to the answersfor the 2 questions Q4 and Q5 The first factor denoted byFactor 1 is the difference in systems and the second factordenoted by Factor 2 is the difference in learning Factor 1includes three levels Edel PPT and our system and Factor2 includes two levels before learning and after learning Theresults from the two-way ANOVA for Q4 shows that therewas no significant interactions between Factor 1 and Factor2 (119865(2 72) = 092 119875 = 040) Furthermore there was asignificant difference for Factor 1 (119865(2 72) = 1926119875 lt 001)while no significant difference existed for Factor 2 (119865(1 72) =158 119875 = 021)The results for Q5 are similar to those for Q4That is there was no significant interaction between Factor1 and Factor 2 (119865(2 72) = 064 119875 = 053) there was asignificant difference for Factor 1 (119865(2 72) = 1400119875 lt 001)and therewas no significant difference for Factor 2 (119865(1 72) =132 119875 = 025)
We then conducted a multiple comparison test for theanswers of the questions Q4 and Q5 we selected Tukeyrsquoshonestly significant difference (HSD) criterion as themultiplecomparison test For Q4 there was a significant differencebetween Edel and our system (119875 lt 001) and there was alsoa significant difference between PPT and our system (119875 lt
001) however there was no significant difference betweenEdel and PPT (see Figure 16) Furthermore for Q5 therewas a significant difference between Edel and our system(119875 lt 001) and therewas also a significant difference betweenPPT and our system (119875 lt 005) however we observed amarginally significant difference between Edel and PPT (119875 lt01) (see Figure 17)
Lastly we applied a one-way ANOVA to the answersfor the question Q6 the factor of this one-way ANOVA isthe difference between systems The results are shown inFigure 18 and the results for the one-way ANOVA show thatthere was a marginally significant difference (119865(2 36) = 259119875 = 0089) We then applied Tukeyrsquos HSD and we observeda marginally significant difference between Edel and oursystem
As a result of the discussion above we can concludethat users feel that our method is easier to produce tactilemaps than the conventional twomethods Furthermore mapimages produced by our system are visually as fine as mapimages produced by Edel and PPT
7 Conclusions
In this paper a computer-aided system for automating pro-duction of tactile graphics from hand-drawn maps has been
0
2
4
6
Edel PPT Our System
Scor
e
Before LearningAfter Learning
lowastlowastlowastlowast
Figure 16 Results for Q4 the average and standard error for eachsystem (lowastlowast119875 lt 001)
0
1
2
3
4
Edel PPT Our System
Scor
e
lowastlowastlowast
Before LearningAfter Learning
Figure 17 Results for Q5 the average and standard error for eachsystem (lowastlowast119875 lt 001 lowast119875 lt 005)
developed The system includes the following four mainprocedures (1) preprocessing (2) segmentation (3) patternrecognition based on fuzzy inference and (4) SVG and Edeldocuments production The preprocessing is applied to aninitial hand-drawn map in order to remove noise and thena binary image is obtained In the segmentation procedurethe binary image is first skeletonized and then the skeletonis segmented into elements by eliminating intersections andcorner points Then a pattern recognition method is appliedto recognize symbols Lastly SVG and Edel documents arecreated to save the recognition results The usability for oursystem was evaluated by comparing our method with the twoconventional systems for creating tactile maps The resultsshow that our system is easier to create tactile maps than theother two systems
In our fuzzy inference systems triangular and trapezoidalmembership functions are applied to define the fuzzy setsdue to their simplicity The shapes of membership functionsinfluence the accuracy of recognition results How to choosean optimal membership function is our future work In thisstudy we do not assume a map includes character stringsHowever adding character strings to a map is very important
Advances in Fuzzy Systems 11
0
1
2
3
4
Edel PPT Our System
Scor
e
Figure 18 Results for Q6 the average and standard error for eachsystem
because it increases the comprehension for the map In thepresent work the types of symbols used in hand-drawnmaps are restricted Developing a system for automatingtranslation of hand-drawn maps with character string andvarious symbols is one of our future works
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This work was supported by the JSPS Grant-in-Aid forScientific Research (C) no 24501198 The authors thank DrKeisuke Ido for giving much advice to the experiments ofusability evaluation
References
[1] J A Miele S Landau and D Gilden ldquoTalking TMAPautomated generation of audio-tactile maps using Smith-Kettlewellrsquos TMAP softwarerdquo The British Journal of VisualImpairment vol 24 no 2 pp 93ndash100 2006
[2] T Watanabe T Yamaguchi K Watanabe et al ldquoDevelopmentand evaluation of a tactile map automated creation systemaccessible to blind personsrdquo IEICE Transactions on Informationand Systems vol J94-D no 10 pp 1652ndash1663 2011 (Japanese)
[3] TMACS httptmacsengniigata-uacjptmacs-dev[4] httptenpuchizugsigojpshokuchizu[5] S E Krufka and K E Barner ldquoAutomatic production of tactile
graphics from scalable vector graphicsrdquo in Proceedings of the 7thInternational ACM SIGACCESS Conference on Computers andAccessibility (ASSETS rsquo05) pp 166ndash172 October 2005
[6] R E Ladner M Y Ivory R Rao et al ldquoAutomating tactilegraphics translationrdquo in Proceedings of the 7th InternationalACMSIGACCESSConference onComputer andAccessibility pp150ndash157 New York NY USA October 2005
[7] Edel httpwww7abiglobenejp EDEL-plus[8] J D Eisenberg SVG Essentials OrsquoREILLY 2002[9] K Broelemann X Jiang and A Schwering ldquoAutomatic street
graph construction in sketchmapsrdquo inGraph-Based Representa-tions in Pattern Recognition X Jiang M Ferrer and A Torsello
Eds vol 6658 of Lecture Notes in Computer Science pp 275ndash284 Springer Berlin Germany 2011
[10] L B Kara and T F Stahovich ldquoAn image-based trainablesymbol recognizer for hand-drawn sketchesrdquo Computers ampGraphics vol 29 no 4 pp 501ndash517 2005
[11] B Edwards and V Chandran ldquoMachine recognition of hand-drawn circuit diagramsrdquo in Proceedings of the IEEE InterntionalConference onAcoustics Speech and Signal Processing vol 6 pp3618ndash3621 June 2000
[12] P Sala ldquoA recognition system for symbols of electronic com-ponents in hand-written circuit diagramsrdquo CSC 2515-MachineLearning Project Report 2004
[13] Y Qiao and M Yasuhara ldquoRecovering dynamic informationfrom static handwritten imagesrdquo in Proceedings of the 9th Inter-national Workshop on Frontiers in Handwriting Recognition pp118ndash123 jpn October 2004
[14] M Notowidigdo and R C Miller ldquoOff-line sketch interpreta-tionrdquo in Proceedings of the AAAI Fall Symposium on MakingPen-Based Interaction Intelligent and Natural pp 120ndash126Arlington Va USA April 2004
[15] R J Gonzalez and R E Woods Digital Image ProcessingPearson Education Upper Saddle River NJ USA 3rd edition2007
[16] N Otsu ldquoA threshold selection method from gray-level his-togramsrdquo IEEE Transactions on Systems Man and Cyberneticsvol 9 no 1 pp 62ndash66 1979
[17] C J Hilditch ldquoLinear skeletons from square cupboardsrdquo inMachine Intelligence Vol IV B Meltzer andDMichie Eds pp403ndash420 Elsevier New York NY USA 1969
[18] S Hermann and R Klette ldquoGlobal curvature estimation for cor-ner detectionrdquo Communication and Information TechnologyResearch Technical Report 171 2005
[19] N Nain V Laxmi B Bhadviya and A Gopal ldquoCorner detec-tion using difference chain code as curvaturerdquo in Proceedingsof the 3rd IEEE International Conference on Signal ImageTechnologies and Internet Based Systems (SITIS 07) pp 821ndash825Shanghai China December 2007
[20] B Kerautret J Lachaud and B Naegel ldquoCurvature basedcorner detector for discrete noisy and multi-scale contoursrdquoInternational Journal of Shape Modeling vol 14 no 2 pp 127ndash145 2008
[21] K Wall and P Danielsson ldquoA fast sequential method forpolygonal approximation of digitized curvesrdquoComputer VisionGraphics amp Image Processing vol 28 no 2 pp 220ndash227 1984
[22] D Chetverikov ldquoA simple and efficient algorithm for detectionof high curvature points in planar curvesrdquo inComputer Analysisof Images and Patterns vol 2756 of Lecture Notes in ComputerScience pp 746ndash753 Springer Berlin Germany 2003
[23] T Terano K Asai and M Sugeno Fuzzy Systems and ItsApplications Academic Press Boston Mass USA 2nd edition1992
[24] N Ono and R Takiyama ldquoOn calculations of curvature ofsampled curvesrdquo Technical Report of IEICE IE93-74 1993(Japanese)
[25] P J Schneider ldquoAn algorithm for automatically fitting digitizedcurvesrdquo in Graphics Gems A S Glassner Ed pp 612ndash625Academic Press 1990
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
8 Advances in Fuzzy Systems
Figure 12 A single cubic Bezier segment
0
2
4
6
8
0h lt 2h 2ndash5h 5ndash10h 10hlt
Figure 13 Experience in computers the vertical axis indicates thenumber of participants
Equation (5) is called a cubic Bezier curve when 119899 = 3Figure 12 shows an example of cubic Bezier curves the fourpoints119881
011988111198812 and119881
3are the control points the thick curve
is the cubic Bezier curve and the polygon expresses the oneconstructed by the four control points
SVG has the formats to express the four shapes andtherefore we can create an SVG document for a hand-drawnmap directly once we have detected the shapes for all theelements of the map On the other hand an Edel documentis a collection of embossed dots So it is easy to create anEdel document once we have detected the shapes for all theelements of a hand-drawn map
5 Experimental Results
This section describes the accuracy for our classificationsystem Five participants drew 15 maps using pencils andpapers These maps were then captured by using an imagescanner the resolution of the scannerwas set to 100 dpiThesedigital images were saved as 24-bits bitmap images The sizesof the images are in 1 169times850 pixelsWe havemeasured theaccuracy for our classification system using precision recalland f -measure
In the 15 maps there are 50 traffic signals 15 destinations15 departure places and 40 landmarks The classificationaccuracy for our system is summarized in Table 1 We canconclude that our system can produce the Edel and SVGdocuments almost correctly from hand-drawn maps
0
2
4
6
8
Freq
uent
ly
Som
etim
es
Occ
asio
nally
Seld
om
Nev
er
Figure 14 Experience in PPT the vertical axis indicates the numberof participants
Table 1 Experimental results are as follows 119873 is the number of allobjects for the symbol119879 is the number of objects correctly classifiedas the symbol 119862 is the number of objects classified as the symbol 119875is precision 119877 is recall and 119865 is 119865-measure
Symbol 119873 119879 119862 119875 () 119877 () 119865 ()Traffic signal 50 50 52 962 100 981Destination 15 15 15 100 100 100Departure place 15 14 14 100 933 965Landmark 40 38 38 100 950 974Cross 130 125 125 100 962 981Arrow 122 121 121 100 992 996Route 15 14 16 875 933 903Railway 28 28 28 100 100 100
6 Usability Evaluation for Our System
61 Method We have studied usability evaluation for oursystemThirteen participants 12 males and 1 female aged 20participated in this investigation all of them are the third-year university students To evaluate the usability for oursystem we selected two common methods for productionof tactile graphics one is the method to use the softwaresystem Edel which assists us to draw diagram images fortactile graphics produced by Braille embossers and anotherone is the method using swell papers In the second methoda diagram image is transferred to a swell paper by a printerand so forth A swell paper has been coated with thermallyfoamed microcapsules that respond to irradiation and causethe dark image on the paper to swell creating the tactilegraphic Microsoft PowerPoint (PPT) was selected to draw adiagram image on a swell paper because this software systemis commonly used in the universities in Japan
Before starting the experiment we asked the followingquestions Q1 Q2 and Q3 to all the participants and theresults for the questionsQ1 andQ3 are summarized in Figures13 and 14 We omit to show the result for the question Q2because all the participants have no experience in using EdelAll the participants are familiar with using computers andhave much experience in using PTT
Advances in Fuzzy Systems 9
(a) (b)
(c) (d)
(e) (f)
Figure 15 Maps for Sessions 1 and 2 (a) Edel map for Session 1 (b) PPT map for Session 1 (c) hand-drawn map for Session 1 (d) Edel mapfor Session 2 (e) PPT map for Session 2 and (f) hand-drawn map for Session 2
Q1 How long do you use computers every weekQ2 Have you ever created figures by using EdelQ3 Have you ever created figures by using PPT
We had two sessions in this investigation For the firstsession a participant created the map diagram shown inFigures 15(a)ndash15(c) the digital files were produced by usingEdel PPT and our system Note that in our system a userfirst draws a map manually using a pencil and paper thenconverts the hand-drawn map to a bitmap image using animage scanner and finally the bitmap image is transformed
into the two digital files the Edel and SVG documents Afterconducting all tasks the participant was asked to answer thefollowing two questions Q4 and Q5
Q4 Were you able to easily create the digital file(s) for thetactile map
Q5 Do you want to use this software system to create thedigital file for a tactile map
After the first session the participant had 5 minutes fortraining the software systems by himselfherself The secondsession followed this learning session In the second session
10 Advances in Fuzzy Systems
the participant conducted similar tasks to the first sessionbut the maps were those shown in Figures 15(d)ndash15(f) Aftercreating all digital files the participant was asked to answertwo questions Q4 and Q5 again and was also asked to answerthe following question Q6
Q6 Were you able to draw the map as you like
The orders of the software systems were determined ran-domly in both the first and the second sessions
62 Results We applied two-way ANOVA to the answersfor the 2 questions Q4 and Q5 The first factor denoted byFactor 1 is the difference in systems and the second factordenoted by Factor 2 is the difference in learning Factor 1includes three levels Edel PPT and our system and Factor2 includes two levels before learning and after learning Theresults from the two-way ANOVA for Q4 shows that therewas no significant interactions between Factor 1 and Factor2 (119865(2 72) = 092 119875 = 040) Furthermore there was asignificant difference for Factor 1 (119865(2 72) = 1926119875 lt 001)while no significant difference existed for Factor 2 (119865(1 72) =158 119875 = 021)The results for Q5 are similar to those for Q4That is there was no significant interaction between Factor1 and Factor 2 (119865(2 72) = 064 119875 = 053) there was asignificant difference for Factor 1 (119865(2 72) = 1400119875 lt 001)and therewas no significant difference for Factor 2 (119865(1 72) =132 119875 = 025)
We then conducted a multiple comparison test for theanswers of the questions Q4 and Q5 we selected Tukeyrsquoshonestly significant difference (HSD) criterion as themultiplecomparison test For Q4 there was a significant differencebetween Edel and our system (119875 lt 001) and there was alsoa significant difference between PPT and our system (119875 lt
001) however there was no significant difference betweenEdel and PPT (see Figure 16) Furthermore for Q5 therewas a significant difference between Edel and our system(119875 lt 001) and therewas also a significant difference betweenPPT and our system (119875 lt 005) however we observed amarginally significant difference between Edel and PPT (119875 lt01) (see Figure 17)
Lastly we applied a one-way ANOVA to the answersfor the question Q6 the factor of this one-way ANOVA isthe difference between systems The results are shown inFigure 18 and the results for the one-way ANOVA show thatthere was a marginally significant difference (119865(2 36) = 259119875 = 0089) We then applied Tukeyrsquos HSD and we observeda marginally significant difference between Edel and oursystem
As a result of the discussion above we can concludethat users feel that our method is easier to produce tactilemaps than the conventional twomethods Furthermore mapimages produced by our system are visually as fine as mapimages produced by Edel and PPT
7 Conclusions
In this paper a computer-aided system for automating pro-duction of tactile graphics from hand-drawn maps has been
0
2
4
6
Edel PPT Our System
Scor
e
Before LearningAfter Learning
lowastlowastlowastlowast
Figure 16 Results for Q4 the average and standard error for eachsystem (lowastlowast119875 lt 001)
0
1
2
3
4
Edel PPT Our System
Scor
e
lowastlowastlowast
Before LearningAfter Learning
Figure 17 Results for Q5 the average and standard error for eachsystem (lowastlowast119875 lt 001 lowast119875 lt 005)
developed The system includes the following four mainprocedures (1) preprocessing (2) segmentation (3) patternrecognition based on fuzzy inference and (4) SVG and Edeldocuments production The preprocessing is applied to aninitial hand-drawn map in order to remove noise and thena binary image is obtained In the segmentation procedurethe binary image is first skeletonized and then the skeletonis segmented into elements by eliminating intersections andcorner points Then a pattern recognition method is appliedto recognize symbols Lastly SVG and Edel documents arecreated to save the recognition results The usability for oursystem was evaluated by comparing our method with the twoconventional systems for creating tactile maps The resultsshow that our system is easier to create tactile maps than theother two systems
In our fuzzy inference systems triangular and trapezoidalmembership functions are applied to define the fuzzy setsdue to their simplicity The shapes of membership functionsinfluence the accuracy of recognition results How to choosean optimal membership function is our future work In thisstudy we do not assume a map includes character stringsHowever adding character strings to a map is very important
Advances in Fuzzy Systems 11
0
1
2
3
4
Edel PPT Our System
Scor
e
Figure 18 Results for Q6 the average and standard error for eachsystem
because it increases the comprehension for the map In thepresent work the types of symbols used in hand-drawnmaps are restricted Developing a system for automatingtranslation of hand-drawn maps with character string andvarious symbols is one of our future works
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This work was supported by the JSPS Grant-in-Aid forScientific Research (C) no 24501198 The authors thank DrKeisuke Ido for giving much advice to the experiments ofusability evaluation
References
[1] J A Miele S Landau and D Gilden ldquoTalking TMAPautomated generation of audio-tactile maps using Smith-Kettlewellrsquos TMAP softwarerdquo The British Journal of VisualImpairment vol 24 no 2 pp 93ndash100 2006
[2] T Watanabe T Yamaguchi K Watanabe et al ldquoDevelopmentand evaluation of a tactile map automated creation systemaccessible to blind personsrdquo IEICE Transactions on Informationand Systems vol J94-D no 10 pp 1652ndash1663 2011 (Japanese)
[3] TMACS httptmacsengniigata-uacjptmacs-dev[4] httptenpuchizugsigojpshokuchizu[5] S E Krufka and K E Barner ldquoAutomatic production of tactile
graphics from scalable vector graphicsrdquo in Proceedings of the 7thInternational ACM SIGACCESS Conference on Computers andAccessibility (ASSETS rsquo05) pp 166ndash172 October 2005
[6] R E Ladner M Y Ivory R Rao et al ldquoAutomating tactilegraphics translationrdquo in Proceedings of the 7th InternationalACMSIGACCESSConference onComputer andAccessibility pp150ndash157 New York NY USA October 2005
[7] Edel httpwww7abiglobenejp EDEL-plus[8] J D Eisenberg SVG Essentials OrsquoREILLY 2002[9] K Broelemann X Jiang and A Schwering ldquoAutomatic street
graph construction in sketchmapsrdquo inGraph-Based Representa-tions in Pattern Recognition X Jiang M Ferrer and A Torsello
Eds vol 6658 of Lecture Notes in Computer Science pp 275ndash284 Springer Berlin Germany 2011
[10] L B Kara and T F Stahovich ldquoAn image-based trainablesymbol recognizer for hand-drawn sketchesrdquo Computers ampGraphics vol 29 no 4 pp 501ndash517 2005
[11] B Edwards and V Chandran ldquoMachine recognition of hand-drawn circuit diagramsrdquo in Proceedings of the IEEE InterntionalConference onAcoustics Speech and Signal Processing vol 6 pp3618ndash3621 June 2000
[12] P Sala ldquoA recognition system for symbols of electronic com-ponents in hand-written circuit diagramsrdquo CSC 2515-MachineLearning Project Report 2004
[13] Y Qiao and M Yasuhara ldquoRecovering dynamic informationfrom static handwritten imagesrdquo in Proceedings of the 9th Inter-national Workshop on Frontiers in Handwriting Recognition pp118ndash123 jpn October 2004
[14] M Notowidigdo and R C Miller ldquoOff-line sketch interpreta-tionrdquo in Proceedings of the AAAI Fall Symposium on MakingPen-Based Interaction Intelligent and Natural pp 120ndash126Arlington Va USA April 2004
[15] R J Gonzalez and R E Woods Digital Image ProcessingPearson Education Upper Saddle River NJ USA 3rd edition2007
[16] N Otsu ldquoA threshold selection method from gray-level his-togramsrdquo IEEE Transactions on Systems Man and Cyberneticsvol 9 no 1 pp 62ndash66 1979
[17] C J Hilditch ldquoLinear skeletons from square cupboardsrdquo inMachine Intelligence Vol IV B Meltzer andDMichie Eds pp403ndash420 Elsevier New York NY USA 1969
[18] S Hermann and R Klette ldquoGlobal curvature estimation for cor-ner detectionrdquo Communication and Information TechnologyResearch Technical Report 171 2005
[19] N Nain V Laxmi B Bhadviya and A Gopal ldquoCorner detec-tion using difference chain code as curvaturerdquo in Proceedingsof the 3rd IEEE International Conference on Signal ImageTechnologies and Internet Based Systems (SITIS 07) pp 821ndash825Shanghai China December 2007
[20] B Kerautret J Lachaud and B Naegel ldquoCurvature basedcorner detector for discrete noisy and multi-scale contoursrdquoInternational Journal of Shape Modeling vol 14 no 2 pp 127ndash145 2008
[21] K Wall and P Danielsson ldquoA fast sequential method forpolygonal approximation of digitized curvesrdquoComputer VisionGraphics amp Image Processing vol 28 no 2 pp 220ndash227 1984
[22] D Chetverikov ldquoA simple and efficient algorithm for detectionof high curvature points in planar curvesrdquo inComputer Analysisof Images and Patterns vol 2756 of Lecture Notes in ComputerScience pp 746ndash753 Springer Berlin Germany 2003
[23] T Terano K Asai and M Sugeno Fuzzy Systems and ItsApplications Academic Press Boston Mass USA 2nd edition1992
[24] N Ono and R Takiyama ldquoOn calculations of curvature ofsampled curvesrdquo Technical Report of IEICE IE93-74 1993(Japanese)
[25] P J Schneider ldquoAn algorithm for automatically fitting digitizedcurvesrdquo in Graphics Gems A S Glassner Ed pp 612ndash625Academic Press 1990
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Advances in Fuzzy Systems 9
(a) (b)
(c) (d)
(e) (f)
Figure 15 Maps for Sessions 1 and 2 (a) Edel map for Session 1 (b) PPT map for Session 1 (c) hand-drawn map for Session 1 (d) Edel mapfor Session 2 (e) PPT map for Session 2 and (f) hand-drawn map for Session 2
Q1 How long do you use computers every weekQ2 Have you ever created figures by using EdelQ3 Have you ever created figures by using PPT
We had two sessions in this investigation For the firstsession a participant created the map diagram shown inFigures 15(a)ndash15(c) the digital files were produced by usingEdel PPT and our system Note that in our system a userfirst draws a map manually using a pencil and paper thenconverts the hand-drawn map to a bitmap image using animage scanner and finally the bitmap image is transformed
into the two digital files the Edel and SVG documents Afterconducting all tasks the participant was asked to answer thefollowing two questions Q4 and Q5
Q4 Were you able to easily create the digital file(s) for thetactile map
Q5 Do you want to use this software system to create thedigital file for a tactile map
After the first session the participant had 5 minutes fortraining the software systems by himselfherself The secondsession followed this learning session In the second session
10 Advances in Fuzzy Systems
the participant conducted similar tasks to the first sessionbut the maps were those shown in Figures 15(d)ndash15(f) Aftercreating all digital files the participant was asked to answertwo questions Q4 and Q5 again and was also asked to answerthe following question Q6
Q6 Were you able to draw the map as you like
The orders of the software systems were determined ran-domly in both the first and the second sessions
62 Results We applied two-way ANOVA to the answersfor the 2 questions Q4 and Q5 The first factor denoted byFactor 1 is the difference in systems and the second factordenoted by Factor 2 is the difference in learning Factor 1includes three levels Edel PPT and our system and Factor2 includes two levels before learning and after learning Theresults from the two-way ANOVA for Q4 shows that therewas no significant interactions between Factor 1 and Factor2 (119865(2 72) = 092 119875 = 040) Furthermore there was asignificant difference for Factor 1 (119865(2 72) = 1926119875 lt 001)while no significant difference existed for Factor 2 (119865(1 72) =158 119875 = 021)The results for Q5 are similar to those for Q4That is there was no significant interaction between Factor1 and Factor 2 (119865(2 72) = 064 119875 = 053) there was asignificant difference for Factor 1 (119865(2 72) = 1400119875 lt 001)and therewas no significant difference for Factor 2 (119865(1 72) =132 119875 = 025)
We then conducted a multiple comparison test for theanswers of the questions Q4 and Q5 we selected Tukeyrsquoshonestly significant difference (HSD) criterion as themultiplecomparison test For Q4 there was a significant differencebetween Edel and our system (119875 lt 001) and there was alsoa significant difference between PPT and our system (119875 lt
001) however there was no significant difference betweenEdel and PPT (see Figure 16) Furthermore for Q5 therewas a significant difference between Edel and our system(119875 lt 001) and therewas also a significant difference betweenPPT and our system (119875 lt 005) however we observed amarginally significant difference between Edel and PPT (119875 lt01) (see Figure 17)
Lastly we applied a one-way ANOVA to the answersfor the question Q6 the factor of this one-way ANOVA isthe difference between systems The results are shown inFigure 18 and the results for the one-way ANOVA show thatthere was a marginally significant difference (119865(2 36) = 259119875 = 0089) We then applied Tukeyrsquos HSD and we observeda marginally significant difference between Edel and oursystem
As a result of the discussion above we can concludethat users feel that our method is easier to produce tactilemaps than the conventional twomethods Furthermore mapimages produced by our system are visually as fine as mapimages produced by Edel and PPT
7 Conclusions
In this paper a computer-aided system for automating pro-duction of tactile graphics from hand-drawn maps has been
0
2
4
6
Edel PPT Our System
Scor
e
Before LearningAfter Learning
lowastlowastlowastlowast
Figure 16 Results for Q4 the average and standard error for eachsystem (lowastlowast119875 lt 001)
0
1
2
3
4
Edel PPT Our System
Scor
e
lowastlowastlowast
Before LearningAfter Learning
Figure 17 Results for Q5 the average and standard error for eachsystem (lowastlowast119875 lt 001 lowast119875 lt 005)
developed The system includes the following four mainprocedures (1) preprocessing (2) segmentation (3) patternrecognition based on fuzzy inference and (4) SVG and Edeldocuments production The preprocessing is applied to aninitial hand-drawn map in order to remove noise and thena binary image is obtained In the segmentation procedurethe binary image is first skeletonized and then the skeletonis segmented into elements by eliminating intersections andcorner points Then a pattern recognition method is appliedto recognize symbols Lastly SVG and Edel documents arecreated to save the recognition results The usability for oursystem was evaluated by comparing our method with the twoconventional systems for creating tactile maps The resultsshow that our system is easier to create tactile maps than theother two systems
In our fuzzy inference systems triangular and trapezoidalmembership functions are applied to define the fuzzy setsdue to their simplicity The shapes of membership functionsinfluence the accuracy of recognition results How to choosean optimal membership function is our future work In thisstudy we do not assume a map includes character stringsHowever adding character strings to a map is very important
Advances in Fuzzy Systems 11
0
1
2
3
4
Edel PPT Our System
Scor
e
Figure 18 Results for Q6 the average and standard error for eachsystem
because it increases the comprehension for the map In thepresent work the types of symbols used in hand-drawnmaps are restricted Developing a system for automatingtranslation of hand-drawn maps with character string andvarious symbols is one of our future works
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This work was supported by the JSPS Grant-in-Aid forScientific Research (C) no 24501198 The authors thank DrKeisuke Ido for giving much advice to the experiments ofusability evaluation
References
[1] J A Miele S Landau and D Gilden ldquoTalking TMAPautomated generation of audio-tactile maps using Smith-Kettlewellrsquos TMAP softwarerdquo The British Journal of VisualImpairment vol 24 no 2 pp 93ndash100 2006
[2] T Watanabe T Yamaguchi K Watanabe et al ldquoDevelopmentand evaluation of a tactile map automated creation systemaccessible to blind personsrdquo IEICE Transactions on Informationand Systems vol J94-D no 10 pp 1652ndash1663 2011 (Japanese)
[3] TMACS httptmacsengniigata-uacjptmacs-dev[4] httptenpuchizugsigojpshokuchizu[5] S E Krufka and K E Barner ldquoAutomatic production of tactile
graphics from scalable vector graphicsrdquo in Proceedings of the 7thInternational ACM SIGACCESS Conference on Computers andAccessibility (ASSETS rsquo05) pp 166ndash172 October 2005
[6] R E Ladner M Y Ivory R Rao et al ldquoAutomating tactilegraphics translationrdquo in Proceedings of the 7th InternationalACMSIGACCESSConference onComputer andAccessibility pp150ndash157 New York NY USA October 2005
[7] Edel httpwww7abiglobenejp EDEL-plus[8] J D Eisenberg SVG Essentials OrsquoREILLY 2002[9] K Broelemann X Jiang and A Schwering ldquoAutomatic street
graph construction in sketchmapsrdquo inGraph-Based Representa-tions in Pattern Recognition X Jiang M Ferrer and A Torsello
Eds vol 6658 of Lecture Notes in Computer Science pp 275ndash284 Springer Berlin Germany 2011
[10] L B Kara and T F Stahovich ldquoAn image-based trainablesymbol recognizer for hand-drawn sketchesrdquo Computers ampGraphics vol 29 no 4 pp 501ndash517 2005
[11] B Edwards and V Chandran ldquoMachine recognition of hand-drawn circuit diagramsrdquo in Proceedings of the IEEE InterntionalConference onAcoustics Speech and Signal Processing vol 6 pp3618ndash3621 June 2000
[12] P Sala ldquoA recognition system for symbols of electronic com-ponents in hand-written circuit diagramsrdquo CSC 2515-MachineLearning Project Report 2004
[13] Y Qiao and M Yasuhara ldquoRecovering dynamic informationfrom static handwritten imagesrdquo in Proceedings of the 9th Inter-national Workshop on Frontiers in Handwriting Recognition pp118ndash123 jpn October 2004
[14] M Notowidigdo and R C Miller ldquoOff-line sketch interpreta-tionrdquo in Proceedings of the AAAI Fall Symposium on MakingPen-Based Interaction Intelligent and Natural pp 120ndash126Arlington Va USA April 2004
[15] R J Gonzalez and R E Woods Digital Image ProcessingPearson Education Upper Saddle River NJ USA 3rd edition2007
[16] N Otsu ldquoA threshold selection method from gray-level his-togramsrdquo IEEE Transactions on Systems Man and Cyberneticsvol 9 no 1 pp 62ndash66 1979
[17] C J Hilditch ldquoLinear skeletons from square cupboardsrdquo inMachine Intelligence Vol IV B Meltzer andDMichie Eds pp403ndash420 Elsevier New York NY USA 1969
[18] S Hermann and R Klette ldquoGlobal curvature estimation for cor-ner detectionrdquo Communication and Information TechnologyResearch Technical Report 171 2005
[19] N Nain V Laxmi B Bhadviya and A Gopal ldquoCorner detec-tion using difference chain code as curvaturerdquo in Proceedingsof the 3rd IEEE International Conference on Signal ImageTechnologies and Internet Based Systems (SITIS 07) pp 821ndash825Shanghai China December 2007
[20] B Kerautret J Lachaud and B Naegel ldquoCurvature basedcorner detector for discrete noisy and multi-scale contoursrdquoInternational Journal of Shape Modeling vol 14 no 2 pp 127ndash145 2008
[21] K Wall and P Danielsson ldquoA fast sequential method forpolygonal approximation of digitized curvesrdquoComputer VisionGraphics amp Image Processing vol 28 no 2 pp 220ndash227 1984
[22] D Chetverikov ldquoA simple and efficient algorithm for detectionof high curvature points in planar curvesrdquo inComputer Analysisof Images and Patterns vol 2756 of Lecture Notes in ComputerScience pp 746ndash753 Springer Berlin Germany 2003
[23] T Terano K Asai and M Sugeno Fuzzy Systems and ItsApplications Academic Press Boston Mass USA 2nd edition1992
[24] N Ono and R Takiyama ldquoOn calculations of curvature ofsampled curvesrdquo Technical Report of IEICE IE93-74 1993(Japanese)
[25] P J Schneider ldquoAn algorithm for automatically fitting digitizedcurvesrdquo in Graphics Gems A S Glassner Ed pp 612ndash625Academic Press 1990
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
10 Advances in Fuzzy Systems
the participant conducted similar tasks to the first sessionbut the maps were those shown in Figures 15(d)ndash15(f) Aftercreating all digital files the participant was asked to answertwo questions Q4 and Q5 again and was also asked to answerthe following question Q6
Q6 Were you able to draw the map as you like
The orders of the software systems were determined ran-domly in both the first and the second sessions
62 Results We applied two-way ANOVA to the answersfor the 2 questions Q4 and Q5 The first factor denoted byFactor 1 is the difference in systems and the second factordenoted by Factor 2 is the difference in learning Factor 1includes three levels Edel PPT and our system and Factor2 includes two levels before learning and after learning Theresults from the two-way ANOVA for Q4 shows that therewas no significant interactions between Factor 1 and Factor2 (119865(2 72) = 092 119875 = 040) Furthermore there was asignificant difference for Factor 1 (119865(2 72) = 1926119875 lt 001)while no significant difference existed for Factor 2 (119865(1 72) =158 119875 = 021)The results for Q5 are similar to those for Q4That is there was no significant interaction between Factor1 and Factor 2 (119865(2 72) = 064 119875 = 053) there was asignificant difference for Factor 1 (119865(2 72) = 1400119875 lt 001)and therewas no significant difference for Factor 2 (119865(1 72) =132 119875 = 025)
We then conducted a multiple comparison test for theanswers of the questions Q4 and Q5 we selected Tukeyrsquoshonestly significant difference (HSD) criterion as themultiplecomparison test For Q4 there was a significant differencebetween Edel and our system (119875 lt 001) and there was alsoa significant difference between PPT and our system (119875 lt
001) however there was no significant difference betweenEdel and PPT (see Figure 16) Furthermore for Q5 therewas a significant difference between Edel and our system(119875 lt 001) and therewas also a significant difference betweenPPT and our system (119875 lt 005) however we observed amarginally significant difference between Edel and PPT (119875 lt01) (see Figure 17)
Lastly we applied a one-way ANOVA to the answersfor the question Q6 the factor of this one-way ANOVA isthe difference between systems The results are shown inFigure 18 and the results for the one-way ANOVA show thatthere was a marginally significant difference (119865(2 36) = 259119875 = 0089) We then applied Tukeyrsquos HSD and we observeda marginally significant difference between Edel and oursystem
As a result of the discussion above we can concludethat users feel that our method is easier to produce tactilemaps than the conventional twomethods Furthermore mapimages produced by our system are visually as fine as mapimages produced by Edel and PPT
7 Conclusions
In this paper a computer-aided system for automating pro-duction of tactile graphics from hand-drawn maps has been
0
2
4
6
Edel PPT Our System
Scor
e
Before LearningAfter Learning
lowastlowastlowastlowast
Figure 16 Results for Q4 the average and standard error for eachsystem (lowastlowast119875 lt 001)
0
1
2
3
4
Edel PPT Our System
Scor
e
lowastlowastlowast
Before LearningAfter Learning
Figure 17 Results for Q5 the average and standard error for eachsystem (lowastlowast119875 lt 001 lowast119875 lt 005)
developed The system includes the following four mainprocedures (1) preprocessing (2) segmentation (3) patternrecognition based on fuzzy inference and (4) SVG and Edeldocuments production The preprocessing is applied to aninitial hand-drawn map in order to remove noise and thena binary image is obtained In the segmentation procedurethe binary image is first skeletonized and then the skeletonis segmented into elements by eliminating intersections andcorner points Then a pattern recognition method is appliedto recognize symbols Lastly SVG and Edel documents arecreated to save the recognition results The usability for oursystem was evaluated by comparing our method with the twoconventional systems for creating tactile maps The resultsshow that our system is easier to create tactile maps than theother two systems
In our fuzzy inference systems triangular and trapezoidalmembership functions are applied to define the fuzzy setsdue to their simplicity The shapes of membership functionsinfluence the accuracy of recognition results How to choosean optimal membership function is our future work In thisstudy we do not assume a map includes character stringsHowever adding character strings to a map is very important
Advances in Fuzzy Systems 11
0
1
2
3
4
Edel PPT Our System
Scor
e
Figure 18 Results for Q6 the average and standard error for eachsystem
because it increases the comprehension for the map In thepresent work the types of symbols used in hand-drawnmaps are restricted Developing a system for automatingtranslation of hand-drawn maps with character string andvarious symbols is one of our future works
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This work was supported by the JSPS Grant-in-Aid forScientific Research (C) no 24501198 The authors thank DrKeisuke Ido for giving much advice to the experiments ofusability evaluation
References
[1] J A Miele S Landau and D Gilden ldquoTalking TMAPautomated generation of audio-tactile maps using Smith-Kettlewellrsquos TMAP softwarerdquo The British Journal of VisualImpairment vol 24 no 2 pp 93ndash100 2006
[2] T Watanabe T Yamaguchi K Watanabe et al ldquoDevelopmentand evaluation of a tactile map automated creation systemaccessible to blind personsrdquo IEICE Transactions on Informationand Systems vol J94-D no 10 pp 1652ndash1663 2011 (Japanese)
[3] TMACS httptmacsengniigata-uacjptmacs-dev[4] httptenpuchizugsigojpshokuchizu[5] S E Krufka and K E Barner ldquoAutomatic production of tactile
graphics from scalable vector graphicsrdquo in Proceedings of the 7thInternational ACM SIGACCESS Conference on Computers andAccessibility (ASSETS rsquo05) pp 166ndash172 October 2005
[6] R E Ladner M Y Ivory R Rao et al ldquoAutomating tactilegraphics translationrdquo in Proceedings of the 7th InternationalACMSIGACCESSConference onComputer andAccessibility pp150ndash157 New York NY USA October 2005
[7] Edel httpwww7abiglobenejp EDEL-plus[8] J D Eisenberg SVG Essentials OrsquoREILLY 2002[9] K Broelemann X Jiang and A Schwering ldquoAutomatic street
graph construction in sketchmapsrdquo inGraph-Based Representa-tions in Pattern Recognition X Jiang M Ferrer and A Torsello
Eds vol 6658 of Lecture Notes in Computer Science pp 275ndash284 Springer Berlin Germany 2011
[10] L B Kara and T F Stahovich ldquoAn image-based trainablesymbol recognizer for hand-drawn sketchesrdquo Computers ampGraphics vol 29 no 4 pp 501ndash517 2005
[11] B Edwards and V Chandran ldquoMachine recognition of hand-drawn circuit diagramsrdquo in Proceedings of the IEEE InterntionalConference onAcoustics Speech and Signal Processing vol 6 pp3618ndash3621 June 2000
[12] P Sala ldquoA recognition system for symbols of electronic com-ponents in hand-written circuit diagramsrdquo CSC 2515-MachineLearning Project Report 2004
[13] Y Qiao and M Yasuhara ldquoRecovering dynamic informationfrom static handwritten imagesrdquo in Proceedings of the 9th Inter-national Workshop on Frontiers in Handwriting Recognition pp118ndash123 jpn October 2004
[14] M Notowidigdo and R C Miller ldquoOff-line sketch interpreta-tionrdquo in Proceedings of the AAAI Fall Symposium on MakingPen-Based Interaction Intelligent and Natural pp 120ndash126Arlington Va USA April 2004
[15] R J Gonzalez and R E Woods Digital Image ProcessingPearson Education Upper Saddle River NJ USA 3rd edition2007
[16] N Otsu ldquoA threshold selection method from gray-level his-togramsrdquo IEEE Transactions on Systems Man and Cyberneticsvol 9 no 1 pp 62ndash66 1979
[17] C J Hilditch ldquoLinear skeletons from square cupboardsrdquo inMachine Intelligence Vol IV B Meltzer andDMichie Eds pp403ndash420 Elsevier New York NY USA 1969
[18] S Hermann and R Klette ldquoGlobal curvature estimation for cor-ner detectionrdquo Communication and Information TechnologyResearch Technical Report 171 2005
[19] N Nain V Laxmi B Bhadviya and A Gopal ldquoCorner detec-tion using difference chain code as curvaturerdquo in Proceedingsof the 3rd IEEE International Conference on Signal ImageTechnologies and Internet Based Systems (SITIS 07) pp 821ndash825Shanghai China December 2007
[20] B Kerautret J Lachaud and B Naegel ldquoCurvature basedcorner detector for discrete noisy and multi-scale contoursrdquoInternational Journal of Shape Modeling vol 14 no 2 pp 127ndash145 2008
[21] K Wall and P Danielsson ldquoA fast sequential method forpolygonal approximation of digitized curvesrdquoComputer VisionGraphics amp Image Processing vol 28 no 2 pp 220ndash227 1984
[22] D Chetverikov ldquoA simple and efficient algorithm for detectionof high curvature points in planar curvesrdquo inComputer Analysisof Images and Patterns vol 2756 of Lecture Notes in ComputerScience pp 746ndash753 Springer Berlin Germany 2003
[23] T Terano K Asai and M Sugeno Fuzzy Systems and ItsApplications Academic Press Boston Mass USA 2nd edition1992
[24] N Ono and R Takiyama ldquoOn calculations of curvature ofsampled curvesrdquo Technical Report of IEICE IE93-74 1993(Japanese)
[25] P J Schneider ldquoAn algorithm for automatically fitting digitizedcurvesrdquo in Graphics Gems A S Glassner Ed pp 612ndash625Academic Press 1990
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Advances in Fuzzy Systems 11
0
1
2
3
4
Edel PPT Our System
Scor
e
Figure 18 Results for Q6 the average and standard error for eachsystem
because it increases the comprehension for the map In thepresent work the types of symbols used in hand-drawnmaps are restricted Developing a system for automatingtranslation of hand-drawn maps with character string andvarious symbols is one of our future works
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This work was supported by the JSPS Grant-in-Aid forScientific Research (C) no 24501198 The authors thank DrKeisuke Ido for giving much advice to the experiments ofusability evaluation
References
[1] J A Miele S Landau and D Gilden ldquoTalking TMAPautomated generation of audio-tactile maps using Smith-Kettlewellrsquos TMAP softwarerdquo The British Journal of VisualImpairment vol 24 no 2 pp 93ndash100 2006
[2] T Watanabe T Yamaguchi K Watanabe et al ldquoDevelopmentand evaluation of a tactile map automated creation systemaccessible to blind personsrdquo IEICE Transactions on Informationand Systems vol J94-D no 10 pp 1652ndash1663 2011 (Japanese)
[3] TMACS httptmacsengniigata-uacjptmacs-dev[4] httptenpuchizugsigojpshokuchizu[5] S E Krufka and K E Barner ldquoAutomatic production of tactile
graphics from scalable vector graphicsrdquo in Proceedings of the 7thInternational ACM SIGACCESS Conference on Computers andAccessibility (ASSETS rsquo05) pp 166ndash172 October 2005
[6] R E Ladner M Y Ivory R Rao et al ldquoAutomating tactilegraphics translationrdquo in Proceedings of the 7th InternationalACMSIGACCESSConference onComputer andAccessibility pp150ndash157 New York NY USA October 2005
[7] Edel httpwww7abiglobenejp EDEL-plus[8] J D Eisenberg SVG Essentials OrsquoREILLY 2002[9] K Broelemann X Jiang and A Schwering ldquoAutomatic street
graph construction in sketchmapsrdquo inGraph-Based Representa-tions in Pattern Recognition X Jiang M Ferrer and A Torsello
Eds vol 6658 of Lecture Notes in Computer Science pp 275ndash284 Springer Berlin Germany 2011
[10] L B Kara and T F Stahovich ldquoAn image-based trainablesymbol recognizer for hand-drawn sketchesrdquo Computers ampGraphics vol 29 no 4 pp 501ndash517 2005
[11] B Edwards and V Chandran ldquoMachine recognition of hand-drawn circuit diagramsrdquo in Proceedings of the IEEE InterntionalConference onAcoustics Speech and Signal Processing vol 6 pp3618ndash3621 June 2000
[12] P Sala ldquoA recognition system for symbols of electronic com-ponents in hand-written circuit diagramsrdquo CSC 2515-MachineLearning Project Report 2004
[13] Y Qiao and M Yasuhara ldquoRecovering dynamic informationfrom static handwritten imagesrdquo in Proceedings of the 9th Inter-national Workshop on Frontiers in Handwriting Recognition pp118ndash123 jpn October 2004
[14] M Notowidigdo and R C Miller ldquoOff-line sketch interpreta-tionrdquo in Proceedings of the AAAI Fall Symposium on MakingPen-Based Interaction Intelligent and Natural pp 120ndash126Arlington Va USA April 2004
[15] R J Gonzalez and R E Woods Digital Image ProcessingPearson Education Upper Saddle River NJ USA 3rd edition2007
[16] N Otsu ldquoA threshold selection method from gray-level his-togramsrdquo IEEE Transactions on Systems Man and Cyberneticsvol 9 no 1 pp 62ndash66 1979
[17] C J Hilditch ldquoLinear skeletons from square cupboardsrdquo inMachine Intelligence Vol IV B Meltzer andDMichie Eds pp403ndash420 Elsevier New York NY USA 1969
[18] S Hermann and R Klette ldquoGlobal curvature estimation for cor-ner detectionrdquo Communication and Information TechnologyResearch Technical Report 171 2005
[19] N Nain V Laxmi B Bhadviya and A Gopal ldquoCorner detec-tion using difference chain code as curvaturerdquo in Proceedingsof the 3rd IEEE International Conference on Signal ImageTechnologies and Internet Based Systems (SITIS 07) pp 821ndash825Shanghai China December 2007
[20] B Kerautret J Lachaud and B Naegel ldquoCurvature basedcorner detector for discrete noisy and multi-scale contoursrdquoInternational Journal of Shape Modeling vol 14 no 2 pp 127ndash145 2008
[21] K Wall and P Danielsson ldquoA fast sequential method forpolygonal approximation of digitized curvesrdquoComputer VisionGraphics amp Image Processing vol 28 no 2 pp 220ndash227 1984
[22] D Chetverikov ldquoA simple and efficient algorithm for detectionof high curvature points in planar curvesrdquo inComputer Analysisof Images and Patterns vol 2756 of Lecture Notes in ComputerScience pp 746ndash753 Springer Berlin Germany 2003
[23] T Terano K Asai and M Sugeno Fuzzy Systems and ItsApplications Academic Press Boston Mass USA 2nd edition1992
[24] N Ono and R Takiyama ldquoOn calculations of curvature ofsampled curvesrdquo Technical Report of IEICE IE93-74 1993(Japanese)
[25] P J Schneider ldquoAn algorithm for automatically fitting digitizedcurvesrdquo in Graphics Gems A S Glassner Ed pp 612ndash625Academic Press 1990
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Submit your manuscripts athttpwwwhindawicom
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Applied Computational Intelligence and Soft Computing
thinspAdvancesthinspinthinsp
Artificial Intelligence
HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation
httpwwwhindawicom Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014