A

Ce

PD

a

ARA

KDLIPD

1

afirabteee[

Catacw[m

r

1h

ARTICLE IN PRESSG ModelEUE-51154; No. of Pages 7

Int. J. Electron. Commun. (AEÜ) xxx (2014) xxx–xxx

Contents lists available at ScienceDirect

International Journal of Electronics andCommunications (AEÜ)

j ourna l h om epage: www.elsev ier .com/ locate /aeue

ontent based image indexing and retrieval using directional localxtrema and magnitude patterns

. Vijaya Bhaskar Reddy ∗, A. Rama Mohan Reddyepartment of Computer Science & Engineering, SVU College of Engineering, Sri Venkateswara University, Tirupati 517502, Andhra Pradesh, India

r t i c l e i n f o

rticle history:eceived 17 June 2013ccepted 23 January 2014

eywords:itectional local extrema patterns (DLEPs)ocal binary patterns (LBPs)

a b s t r a c t

In this paper, we integrate the concept of directional local extremas and their magnitude based patternsfor content based image indexing and retrieval. The standard ditectional local extrama pattern (DLEP)extracts the directional edge information based on local extrema in 0◦, 45◦, 90◦, and 135◦ directions inan image. However, they are not considering the magnitudes of local extremas. The proposed methodintegrates these two concepts for better retrieval performance. The sign DLEP (SDLEP) operator is ageneralized DLEP operator and magnitude DLEP (MDLEP) operator is calculated using magnitudes of

mage retrievalattern recognitionatabases

local extremas. The performance of the proposed method is compared with DLEP, local binary patterns(LBPs), block-based LBP (BLK LBP), center-symmetric local binary pattern (CS-LBP), local edge patternsfor segmentation (LEPSEG) and local edge patterns for image retrieval (LEPINV) methods by conductingtwo experiments on benchmark databases, viz. Corel-5K and Corel-10K databases. The results after beinginvestigated show a significant improvement in terms of their evaluation measures as compared to otherexisting methods on respective databases.

. Introduction

Retrieval of images from large image databases has been anctive area of research for long due to its applications in variouselds like satellite imaging, medicine, etc. Content based imageetrieval (CBIR) systems extract features from the raw imagesnd calculate an associative measure (similarity or dissimilarity)etween a query image and database images based on these fea-ures. Hence the feature extraction is a very important step and theffectiveness of a CBIR system depends typically on the method ofxtraction of features from raw images. Several methods achievingffective feature extraction have been proposed in the literature1–4].

Texture is the most important feature for CBIR. Smith andhang used the mean and variance of the wavelet coefficientss texture features for CBIR [5]. Moghaddam et al. proposedhe Gabor wavelet correlogram (GWC) for CBIR [6,7]. Ahmadiannd Mostafa used the wavelet transform for texture classifi-ation [8]. Moghaddam et al. Introduced new algorithm called

Please cite this article in press as: Vijaya Bhaskar Reddy P, Rama Modirectional local extrema and magnitude patterns. Int J Electron Comm

avelet correlogram (WC) [9]. Saadatmand and Moghaddam7,10] improved the performance of the WC algorithm by opti-

izing the quantization thresholds using genetic algorithm (GA).

∗ Corresponding author. Tel.: +91 9666508044.E-mail addresses: pvbreddy6@gmail.com (P. Vijaya Bhaskar Reddy),

amamohansvu@yahoo.com (A. Rama Mohan Reddy).

434-8411/$ – see front matter © 2014 Elsevier GmbH. All rights reserved.ttp://dx.doi.org/10.1016/j.aeue.2014.01.012

© 2014 Elsevier GmbH. All rights reserved.

Birgale et al. [11] and Subrahmanyam et al. [12] combined thecolor (color histogram) and texture (wavelet transform) featuresfor CBIR. Subrahmanyam et al. proposed correlogram algorithm forimage retrieval using wavelets and rotated wavelets (WC + RWC)[13].

Ojala et al. proposed the local binary pattern (LBP) features fortexture description [14] and these LBPs are converted to rotationalinvariant for texture classification [15]. Pietikainen et al. proposedthe rotational invariant texture classification using feature distri-butions [16]. Ahonen et al. [17] and Zhao and Pietikainen [18]used the LBP operator facial expression analysis and recognition.Heikkila and Pietikainen proposed the background modeling anddetection by using LBP [19]. Huang et al. proposed the extendedLBP for shape localization [20]. Heikkila et al. used the LBP forinterest region description [21]. Li and Staunton used the combi-nation of Gabor filter and LBP for texture segmentation [22]. Zhanget al. proposed the local derivative pattern for face recognition [23].They have considered LBP as a nondirectional first order local pat-tern, which are the binary results of the first-order derivative inimages.

The block-based texture feature which use the LBP texture fea-ture as the source of image description is proposed in [24] forCBIR. The center-symmetric local binary pattern (CS-LBP) which

han Reddy A. Content based image indexing and retrieval usingun (AEÜ) (2014), http://dx.doi.org/10.1016/j.aeue.2014.01.012

is a modified version of the well-known LBP feature is com-bined with scale invariant feature transform (SIFT) in [25] fordescription of interest regions. Yao and Chen [26] have pro-posed two types of local edge patterns (LEP) histograms, one is

ING ModelA

2 / Int.

Lrosc

ouDwto

otpS4i

2

2

tpson

ARTICLEEUE-51154; No. of Pages 7

P. Vijaya Bhaskar Reddy, A. Rama Mohan Reddy

EPSEG for image segmentation, and the other is LEPINV for imageetrieval. The LEPSEG is sensitive to variations in rotation and scale,n the contrary, the LEPINV is resistant to variations in rotation andcale. Subrahmanyam et al. [27] have proposed the DLEP whichollects the directional edge information for image retrieval.

The above discussed various extensions of LBP features considernly the sign of differences but not magnitudes. The main contrib-tions of this work are summarized as follows: (a) the existingLEPs are considering only sign of difference between the pixelshereas our method considers the both sign as well as magni-

udes and (b) the performance of the proposed method is testedn benchmark image databases.

The paper is summarized as follows: in Section 1, a brief reviewf content based image retrieval and related work is given. Sec-ion 2, presents a concise review of local pattern operators. Theroposed system framework and query matching are illustrated inection 3. Experimental results and discussions are given in Section. Based on above work, conclusions and future scope are derived

n Section 5.

. Local patterns

.1. Local binary patterns (LBPs)

The LBP operator was introduced by Ojala et al. [14] for tex-ure classification. Success in terms of speed (no need to tune any

Please cite this article in press as: Vijaya Bhaskar Reddy P, Rama Modirectional local extrema and magnitude patterns. Int J Electron Comm

arameters) and performance is reported in many research areasuch as texture classification [14–16], face recognition [17,18],bject tracking, bio-medical image retrieval and finger print recog-ition. Given a center pixel in the 3 × 3 pattern, LBP value is

Fig. 1. Calculati

Fig. 2. Proposed image retrie

PRESSJ. Electron. Commun. (AEÜ) xxx (2014) xxx–xxx

computed by comparing its gray scale value with its neighborhoodsbased on Eq. (1) and (2):

LBPP,R =P∑

p=1

2(p−1) × f1(I(gp) − I(gc)) (1)

f1(x) ={

1 x≥0

0 else(2)

where I(gc) denotes the gray value of the center pixel, I(gp) repre-sents the gray value of its neighbors, P stands for the number ofneighbors and R, the radius of the neighborhood.

After computing the LBP pattern for each pixel (j, k), the wholeimage is represented by building a histogram as shown in Eq. (3).

HLBP(l) =N1∑j=1

N2∑k=1

f2(LBP(j, k), l); l ∈ [0, (2P − 1)] (3)

f2(x, y) ={

1 x = y

0 else(4)

where the size of input image is N1 × N2.Fig. 1 shows an example of obtaining an LBP from a given 3 × 3

pattern. The histograms of these patterns contain the informationon the distribution of edges in an image.

2.2. Block based local binary patterns (BLK LBP)

han Reddy A. Content based image indexing and retrieval usingun (AEÜ) (2014), http://dx.doi.org/10.1016/j.aeue.2014.01.012

Takala et al. [24] have proposed the block based LBP for CBIR. Theblock division method is a simple approach that relies on subimagesto address the spatial properties of images. It can be used togetherwith any histogram descriptors similar to LBP. The method works

on of LBP.

val system framework.

IN PRESSG ModelA

/ Int. J. Electron. Commun. (AEÜ) xxx (2014) xxx–xxx 3

ibctt

2

eC

C

AiL

2

(t

9b

I

T

I

f

T

v

b

H

w

nswmrfa

2

vi

Fig. 3. Comparison of proposed method with other existing methods on Corel-5K.(a) Category wise performance in terms of precision, (b) category wise performance

ARTICLEEUE-51154; No. of Pages 7

P. Vijaya Bhaskar Reddy, A. Rama Mohan Reddy

n the following way: First it divides the model images into squarelocks that are arbitrary in size and overlap. Then the method cal-ulates the LBP distributions for each of the blocks and combineshe histograms into a single vector of sub-histograms representinghe image.

.3. Center-symmetric local binary patterns (CS LBP)

Instead of comparing each pixel with the center pixel, Heikkilat al. [25] have compared center-symmetric pairs of pixels forS LBP as shown in Eq. (5):

S LBPP,R =P∑

p=1

2(p−1) × f1(I(gp) − I(gp+(P/2))) (5)

fter computing the CS LBP pattern for each pixel (j, k), the wholemage is represented by building a histogram as similar to theBP.

.4. Directional local extrema patterns (DLEPs)

Subrahmanyam et al. [27] directional local extrema patternsDLEPs) for CBIR. DLEP describes the spatial structure of the localexture using the local extrema of center gray pixel gc .

In proposed DLEP for a given image the local extrema in 0◦, 45◦,0◦, and 135◦ directions are obtained by computing local differenceetween the center pixel and its neighbors as shown below:

′(gi) = I(gc) − I(gi); i = 1, 2, . . ., 8 (6)

he local extremas are obtained by Eq. (7).

˛(gc) = f3(I′(gj), I′(gj+4)); j = 1 + ˛

45∀ = 0◦, (7)

3(I′(gj), I′(gj+4)) ={

1 I′(gj) × I′(gj+4)≥0

0 else(8)

he DLEP is defined ( = 0◦, 45◦, 90◦, and 135◦) as follows:

DLEP(I(gc))∣∣˛

= {I˛(gc); I˛(g1); I˛(g2); . . .I˛(g8)} (9)

Eventually, the given image is converted to DLEP images withalues ranging from 0 to 511.

After calculation of DLEP, the whole image is represented byuilding a histogram supported by Eq. (10) [27].

DLEP|˛ (l) =N1∑j=1

N2∑k=1

f2( DLEP(j, k)∣∣˛

, l); l ∈ [0, 511] (10)

here the size of input image is N1 × N2.In literature [28] and [29], it is already proved that the mag-

itude of the difference patterns along with sign patterns show aignificant improvement in the retrieval performance as comparedith the sign patterns. The concept which is available in [28,29] isotivated us to propose the magnitude DLEP patterns for image

etrieval. In this paper, we combine the DLEP and magnitude DLEPeatures for image retrieval and shows a significant improvements compared to the DLEP alone (see Section 4).

.5. Magnitude directional local extrema patterns (MDLEPs)

Please cite this article in press as: Vijaya Bhaskar Reddy P, Rama Modirectional local extrema and magnitude patterns. Int J Electron Comm

The existing DLEP [27] considers only the sign of local extremaalues which are calculated between the given center pixel andts surrounding neighbors. From the above observation it can be

in terms of recall, (c) total database performance in terms of average precision and(d) total database performance in terms of ARR.

analyze that there is a possible to increase the performance of thesystem by considering the magnitude of local extremas.

The magnitude patterns for local extremas are calculated fol-lows.

IM˛ (gc) = f4(I′(gj), I′(gj+4)); j = 1 + ˛

45∀ = 0◦, 45◦, 90◦, and 135◦

(11)

{

han Reddy A. Content based image indexing and retrieval usingun (AEÜ) (2014), http://dx.doi.org/10.1016/j.aeue.2014.01.012

f4(I′(gj), I′(gj+4)) =1 abs(I′(gj)) + abs(I′(gj+4))≥Th

0 else(12)

ARTICLE IN PRESSG ModelAEUE-51154; No. of Pages 7

4 P. Vijaya Bhaskar Reddy, A. Rama Mohan Reddy / Int. J. Electron. Commun. (AEÜ) xxx (2014) xxx–xxx

Table 1Results of various methods in terms of precision and recall on Corel-5K and Corel-10K databases PM: DLEP + MDLEP; BLK LBP: block based LBP [24].

Database Performance Method

CS LBP LEPSEG LEPINV BLK LBP LBP DLEP PM

Corel-5KPrecision (%) 32.9 41.5 35.1 45.7 43.6 48.8 54.4Recall (%) 14.0 18.3 14.8 20.3 19.2 21.1 24.1

Corel-10KPrecision (%) 26.4 34.0 28.9 38.1 37.6 40.0 45.4Recall (%) 10.1 13.8 11.2 15.3 14.9 15.7 18.4

T

T

Ab

h = 1N1 × N2

N1∑i=1

N2∑k=1

(abs( I′(gj)∣∣(i,k)

) + abs( I′(gj+4)∣∣(i,k)

)) (13)

he MDLEP is defined ( = 0◦, 45◦, 90◦, and 135◦) as follows:

Please cite this article in press as: Vijaya Bhaskar Reddy P, Rama Modirectional local extrema and magnitude patterns. Int J Electron Comm

MDLEP(I(gc))∣∣˛

= {IM˛ (gc); IM

˛ (g1); IM˛ (g2); .. . .IM

˛ (g8)} (14)

fter calculation of MDLEP, the whole image is represented byuilding a histogram supported by Eq. (10).

Fig. 4. Two examples of image retrieval by propo

3. Proposed system framework

3.1. Image retrieval system

In this paper, we integrate the features of DLEP and magnitude

han Reddy A. Content based image indexing and retrieval usingun (AEÜ) (2014), http://dx.doi.org/10.1016/j.aeue.2014.01.012

DLEP for image retrieval. First, the image is loaded and convertedinto gray scale if it is RGB. Secondly, the DLEPs and magnitudeDLEPs (MDLEPs) are collected and then go for the histograms calcu-lation. Finally, the feature vector is generated by concatenating the

sed method (DLEP) on Corel-5K database.

IN PRESSG ModelA

/ Int. J. Electron. Commun. (AEÜ) xxx (2014) xxx–xxx 5

hp

A

I123

4

567

3

(ef

boiw

D

w

4

dm

biigtac

qXpshf

or

p

a

r

Fig. 5. Comparison of proposed method with other existing methods on Corel-10K.

formance of the proposed method is measured in terms of ARP andARR as shown in Eqs. (16)–(19).

Table 1 illustrates the retrieval results of proposed method andother existing methods on Corel-5K and Corel-10K databases in

ARTICLEEUE-51154; No. of Pages 7

P. Vijaya Bhaskar Reddy, A. Rama Mohan Reddy

istograms of DLEP and MDLEP. Fig. 2 depicts the flowchart of theroposed technique and algorithm for the same is presented here:

lgorithm.

nput: Image; Output: Retrieval result. Load the image and converted in to gray scale (if it is RGB).. Calculate the local extrema in 0◦ , 45◦ , 90◦ , and 135◦ directions.. Compute the DLEP and MDLEP patterns in 0◦ , 45◦ , 90◦ , and 135◦

directions.. Construct the histograms for DLEP and MDLEP patterns in 0◦ , 45◦ , 90◦ ,

and 135◦ directions.. Construct the feature vector by concatenating all histograms.. Compare the query image with the image in the database using Eq. (15).. Retrieve the images based on the best matches.

.2. Query matching

Feature vector for query image Q is represented as fQ =fQ1 , fQ2 , . . .fQLg

) obtained after the feature extraction. Similarlyach image in the database is represented with feature vector

DBj= (fDBj1

, fDBj2, . . .fDBjLg

); j = 1, 2, . . .,∣∣DB

∣∣. The goal is to select nest images that resemble the query image. This involves selectionf n top matched images by measuring the distance between querymage and image in the database

∣∣DB∣∣. In order to match the images

e used d1 similarity distance metric [27] computed by Eq. (15).

(Q, DB) =Lg∑

i=1

∣∣∣∣ fDBji− fQi

1 + fDBji+ fQi

∣∣∣∣ (15)

here fDBjiis ith feature of jth image in the database

∣∣DB∣∣.

. Experiments

The effectiveness of the proposed method is analyzed by con-ucting two experiments on benchmark databases. Further, it isentioned that the databases used are Corel-5K and Corel-10K.In experiments #1 and #2, images from Corel database [30] have

een used. This database consists of large number of images of var-ous contents ranging from animals to outdoor sports to naturalmages. These images have been pre-classified into different cate-ories each of size 100 by domain professionals. Some researchershink that Corel database meets all the requirements to evaluaten image retrieval system, due its large size and heterogeneousontent.

In all experiments, each image in the database is used as theuery image. For each query, the system collects n database images

= (x1, x2, . . ., xn) with the shortest image matching distance com-uted using Eq. (15). If the retrieved image xi = 1, 2, . . ., n belongs toame category as that of the query image then we say the systemas appropriately identified the expected image else the system

ails to find the expected image.The performance of the proposed method is measured in terms

f average precision/average retrieval precision (ARP), averageecall/average retrieval rate (ARR) as shown below:

For the query image Iq, the precision is defined as follows:

recision, P(Iq) = number of relevant images retrievedtotal number of images retrieved

(16)

verage precision, ARP = 1∣∣DB∣∣

|DB|∑i=1

P(Ii)

∣∣∣∣∣∣ (17)

ecall, R(Iq)

= number of relevant images retrievedtotal number of relevant images in the database

(18)

Please cite this article in press as: Vijaya Bhaskar Reddy P, Rama Modirectional local extrema and magnitude patterns. Int J Electron Comm

(a) Category wise performance in terms of precision, (b) category wise performancein terms of recall, (c) total database performance in terms of average precision and(d) total database performance in terms of ARR.

average recall, ARR = 1∣∣DB∣∣

|DB|∑i=1

R(Ii)

∣∣∣∣∣∣ (19)

4.1. Corel-5K database

Corel-5K database consists of 5000 images which are collectedfrom 50 different domains have 100 images per domain. The per-

han Reddy A. Content based image indexing and retrieval usingun (AEÜ) (2014), http://dx.doi.org/10.1016/j.aeue.2014.01.012

ARTICLE IN PRESSG ModelAEUE-51154; No. of Pages 7

6 P. Vijaya Bhaskar Reddy, A. Rama Mohan Reddy / Int. J. Electron. Commun. (AEÜ) xxx (2014) xxx–xxx

proposed method (DLEP) on Corel-10K database.

tcrts

1

2

ptCm

4

ffpi

opAts

Table 2Computational cost of various methods for the feature extraction time and retrievaltime for a given query image.

Method Feature extraction time (s) Retrieval time (s) Total

LBP 0.03 0.02 0.05CS LBP 0.04 0.015 0.055LEPSEG 0.10 0.02 0.12LEPINV 0.15 0.02 0.17DLEP 0.21 0.026 0.236

Fig. 6. Two examples of image retrieval by

erms of average precision and recall. Fig. 3(a) and (b) shows theategory wise performance of methods in terms of precision andecall on Corel-5K database. The performance of all techniques inerms of average precision and ARR on Corel-5K database can beeen in Fig. 3(c) and (d), respectively.

From Table 1 and Fig. 3, the following points are observed.

. The proposed method (DLEP + MDLEP) showing 21.5%, 12.9%,19.3%, 8.7%, 10.8% and 5.6% more performance as compared toCS LBP, LEPSEG, LEPINV, BLK LBP, LBP and DLEP, respectively, interms of ARP on Core-5K database.

. The proposed method (DLEP + MDLEP) showing 10.1%, 5.8%, 9.3%,3.8%, 4.9% and 3% more performance as compared to CS LBP, LEP-SEG, LEPINV, BLK LBP, LBP and DLEP, respectively, in terms ofARR on Core-5K database.

From Table 1, Fig. 3 and above observations, it is clear that theroposed method shows a significant improvement as comparedo other existing methods in terms of their evaluation measures onorel-5K database. Fig. 4 illustrates the query results of proposedethod on Corel-5K database (top left image is the query image).

.2. Corel-10K database

Corel-5K database consists of 10,000 images which are collectedrom 100 different domains have 100 images per domain. The per-ormance of the proposed method is measured in terms of averagerecision, average recall, and average retrieval rate (ARR) as shown

n Eqs. (16)–(19) (Fig. 4).Fig. 5(a) and (b) shows the category wise performance of meth-

ds in terms of precision and recall on Corel-10K database. The

Please cite this article in press as: Vijaya Bhaskar Reddy P, Rama Modirectional local extrema and magnitude patterns. Int J Electron Comm

erformance of all techniques in terms of average precision andRR on Corel-10K database can be seen in Fig. 5(c) and (d), respec-

ively. From Table 1 and Fig. 5, it is clear that the proposed methodhows a significant improvement as compared to other existing

PM 0.25 0.027 0.277

Query image size is 256 × 384.

methods in terms of their evaluation measures on Corel-10Kdatabase. Fig. 6 illustrates the query results of proposed methodon Corel-10K database (top left image is the query image).

4.3. Computational cost vs. performance

Table 2 illustrates the computational cost of various methodsfor the feature extraction time and retrieval time for a given queryimage. The computational cost of the proposed method is little bitmore as compared to the DLEP, as it outperforms:

1. The DLEP by 5.6% in terms of ARP on Corel-5K database.2. The DLEP by 3.0% in terms of ARR on Corel-5K database.3. The DLEP by 5.4% in terms of ARP on Corel-10K database.4. The DLEP by 2.7% in terms of ARR on Corel-10K database.

From the above observations, it is clear that the proposedmethod shows a significant improvement in retrieval performancewith the small increment of computational cost.

han Reddy A. Content based image indexing and retrieval usingun (AEÜ) (2014), http://dx.doi.org/10.1016/j.aeue.2014.01.012

5. Conclusions

A new approach which integrates the DLEP and MDLEP featuresfor content based image retrieval is presented in this paper. The

ING ModelA

/ Int.

pitiirtd

A

lca

R

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

Tirupati, Andhra Pradesh, India. He has more than 20 yearsof teaching experience. He published many papers in thepeer-refereed journals and conferences. His major fields

ARTICLEEUE-51154; No. of Pages 7

P. Vijaya Bhaskar Reddy, A. Rama Mohan Reddy

roposed MDLEP differs from the existing DLEP in a manner thatt extracts the directional edge information based on the magni-udes of local extrema in 0◦, 45◦, 90◦, and 135◦ directions in anmage. Performance of the proposed method is tested by conduct-ng two experiments on benchmark image databases and retrievalesults show a significant improvement in terms of their evalua-ion measures as compared to other existing methods on respectiveatabases.

cknowledgment

Our sincere thanks to Dr. Subrahmanyam Murala, Post-Doc Fel-ow, University of Windsor, ON, Canada for providing the sourceode and results for DLEP and LBP variant methods on Corel-5Knd Corel-10K databases.

eferences

[1] Rui Y, Huang TS. Image retrieval: current techniques, promising directions andopen issues. J Vis Commun Image Represent 1999;10:39–62.

[2] Smeulders AWM, Worring M, Santini S, Gupta A, Jain R. Content-based imageretrieval at the end of the early years. IEEE Trans Pattern Anal Mach Intell2000;22(12):1349–80.

[3] Kokare M, Chatterji BN, Biswas PK. A survey on current content based imageretrieval methods. IETE J Res 2002;48(3/4):261–71.

[4] Liu Y, Zhang D, Lu G, Ma W-Y. A survey of content-based image retrieval withhigh-level semantics. Pattern Recogn 2007;40:262–82.

[5] Smith JR, Chang SF. Automated binary texture feature sets for image retrieval.In: Proc. IEEE int. conf. acoustics, speech and signal processing. New York:Columbia Univ.; 1996. p. 2239–42.

[6] Moghaddam HA, Khajoie TT, Rouhi AH. A new algorithm for image indexingand retrieval using wavelet correlogram. In: Int. conf. image processing, vol. 2.Tehran, Iran: K.N. Toosi Univ. of Technol.; 2003. p. 497–500.

[7] Saadatmand MT, Moghaddam HA. Enhanced wavelet correlogram methods forimage indexing and retrieval. In: IEEE int. conf. image processing. Tehran, Iran:K.N. Toosi Univ. of Technol.; 2005. p. 541–4.

[8] Ahmadian A, Mostafa A. An Efficient Texture Classification Algorithm usingGabor wavelet. In: 25th annual international conf. of the IEEE EMBS. 2003. p.930–3.

[9] Moghaddam HA, Khajoie TT, Rouhi AH, Saadatmand MT. Wavelet correl-ogram: a new approach for image indexing and retrieval. Pattern Recogn2005;38(12):2506–18.

10] Saadatmand MT, Moghaddam HA. A novel evolutionary approach foroptimizing content based image retrieval. IEEE Trans Syst Man Cybern2007;37(1):139–53.

11] Birgale L, Kokare M, Doye D. Color and texture features for content based imageretrieval. In: International conf. computer graphics, image and visualisation.2006. p. 146–9.

12] Subrahmanyam M, Gonde AB, Maheshwari RP. Color and texture features forimage indexing and retrieval. In: IEEE int. advance computing conf. 2009. p.1411–6.

13] Subrahmanyam M, Maheshwari RP, Balasubramanian R. A correlogram algo-rithm for image indexing and retrieval using wavelet and rotated waveletfilters. Int J Signal Imaging Syst Eng 2011;4(1):27–34.

14] Ojala T, Pietikainen M, Harwood D. A comparative study of texture meas-ures with classification based on feature distributions. J Pattern Recogn1996;29(1):51–9.

15] Ojala T, Pietikainen M, Maenpaa T. Multiresolution gray-scale and rotation

Please cite this article in press as: Vijaya Bhaskar Reddy P, Rama Modirectional local extrema and magnitude patterns. Int J Electron Comm

invariant texture classification with local binary patterns. IEEE Trans PatternAnal Mach Intell 2002;24(7):971–87.

16] Pietikainen M, Ojala T, Scruggs T, Bowyer KW, Jin C, Hoffman K, et al.Overview of the face recognition using feature distributions. Pattern Recogn2000;33(1):43–52.

PRESSJ. Electron. Commun. (AEÜ) xxx (2014) xxx–xxx 7

17] Ahonen T, Hadid A, Pietikainen M. Face description with local binary pat-terns: applications to face recognition. IEEE Trans Pattern Anal Mach Intell2006;28(12):2037–41.

18] Zhao G, Pietikainen M. Dynamic texture recognition using local binary patternswith an application to facial expressions. IEEE Trans Pattern Anal Mach Intell2007;29(6):915–28.

19] Heikkil MA, Pietikainen M. A texture based method for modeling the back-ground and detecting moving objects. IEEE Trans Pattern Anal Mach Intell2006;28(4):657–62.

20] Huang X, Li SZ, Wang Y. Shape localization based on statistical method usingextended local binary patterns. In: Proc. int. conf. image graphics. 2004. p.184–7.

21] Heikkila M, Pietikainen M, Schmid C. Description of interest regions with localbinary patterns. Pattern Recogn 2009;42:425–36.

22] Li M, Staunton RC. Optimum Gabor filter design and local binary patterns fortexture segmentation. Pattern Recogn 2008;29:664–72.

23] Zhang B, Gao Y, Zhao S, Liu J. Local derivative pattern versus local binary pattern:face recognition with higher-order local pattern descriptor. IEEE Trans ImageProc 2010;19(2):533–44.

24] Takala V, Ahonen T, Pietikainen M. Block-based methods for imageretrieval using local binary patterns. In: SCIA 2005, vol. 3450. 2005.p. 882–91.

25] Heikkil M, Pietikainen M, Schmid C. Description of interest regions with localbinary patterns. Pattern Recogn 2009;42:425–36.

26] Yao C-H, Chen S-Y. Retrieval of translated, rotated and scaled color textures.Pattern Recogn 2003;36:913–29.

27] Subrahmanyam M, Maheshwari RP, Balasubramanian R. Directional localextrema patterns: a new descriptor for content based image retrieval. Int JMultimedia Inform Retrieval 2012;1(3):191–203.

28] Subrahmanyam M, Maheshwari RP, Balasubramanian R. Sign and magni-tude patterns for image indexing and retrieval. Int J Comput Vis Robot2010;1(3):279–96.

29] Subrahmanyam M, Maheshwari RP, Balasubramanian R. Local tetra patterns:a new feature descriptor for content based image retrieval. IEEE Trans ImageProcess 2012;21(5):2874–86.

30] Corel-10K image database. Available: http://www.ci.gxnu.edu.cn/cbir/Dataset.aspx

P. Vijaya Bhaskar Reddy has completed his M. Tech inComputer Science and Engineering from Bharath Univer-sity, Chennai in 2008. Currently he is pursuing the Ph.D.degree in the Department of Computer Science and Engi-neering, Sri Venkateswara University, Tirupati, AndhraPradesh, India. His major fields of interests are ContentBased Image Retrieval, Image Processing and PatternsRecognition.

A. Rama Mohan Reddy is working as Professor & HODat Department of Computer Science and Engineering, SriVenkateswara University, Tirupati, Andhra Pradesh, India.He Completed his M. Tech. in Computer Science from NITWarangal. After that, he completed his Ph.D. in the area ofsoftware Architecture from Sri Venkateswara University,

han Reddy A. Content based image indexing and retrieval usingun (AEÜ) (2014), http://dx.doi.org/10.1016/j.aeue.2014.01.012

of interests are Image Processing, Software Engineering,Software Architecture and Data Mining.