hand veins recognition system - fenix.tecnico.ulisboa.pt · hand-based biometrics, biometrics...

Hand Veins Recognition System

João Ricardo Gonçalves Neves

Thesis to obtain the Master of Science Degree in

Electrical and Computer Engineering

Examination Committee

Chairperson: Prof. Fernando Duarte Nunes

Supervisor: Prof. Paulo Luís Serras Lobato Correia

Members of the Committee: Prof. Hugo Pedro Martins Carriço Proença

September 2013

Hand Veins Recognition System 2013

i João Ricardo Gonçalves Neves

Acknowledgements

In first place, I would like to thank Professor Paulo Lobato Correia for his support,

availability, ideas and continuous guidance provided throughout the development

of this dissertation.

I also want to thank Nuno Moço for all his assistance in the programming area, and

Pedro Fernandes for his support in the conception and construction of the

prototype used in the thesis.

Without the contribution of my family and friends, it would have been impossible

to have so many hand veins acquisitions available in the database, and for that I am

really thankful.

I want to express my gratitude to my family, for the all the patience, guidance and

support provided in the last years: To my father who was always available to

provide new useful ideas that might improve my thesis; to my mother, who was

always available to listen to my complaints and to give advice; and to my brother

who was always there to provide his help.

I also want to thank my girlfriend Mariana Cadete for all her motivation, support

and patience through all the superior education years.


ii João Ricardo Gonçalves Neves

Abstract

Accurate protection systems capable of replacing the traditional passwords and ID

cards are essential, for commodity and for security reasons. A hand-vein pattern

recognition system is just one of a vast group of biometric techniques under

research, in order to become the reference recognition system.

This dissertation presents a hand vein biometric recognition system that uses the

hand blood vessels pattern to identify an individual. All biometric systems have an

immense application potential, as they present advantages over the traditional

identification systems. They are able to work with patterns that are very hard to

duplicate, since they are different from person to person, and it is also impossible

to lose or forget them, since the biometric characteristics are intrinsically attached

to the human body.

The developed approach was created with the intent of providing an effective

protection system despite having been designed and implemented using

inexpensive hardware, in comparison with the biometric recognition systems

presently offered at a commercial level.

The results show that a reliable system can be produced at a low cost and can be

used standalone or in combination with other systems.

Keywords

Hand-based biometrics, biometrics recognition, palm veins, hand geometry, palm vein

acquisition system, palmprint, web-camera.


iii João Ricardo Gonçalves Neves

Resumo

A necessidade de sistemas de identificação eficientes e de baixo custo aumenta à

medida que as transações de dados pessoais ou valores por via eletrónica também

aumentam. O sistema de reconhecimento de padrões de veias desenvolvido nesta

dissertação é apenas um de um vasto grupo de técnicas de reconhecimento

biométrico que estão a ser exploradas presentemente. Estes sistemas estão a ser

desenvolvidos com o objetivo de se tornarem os métodos de proteção de

referência, superiores aos métodos de proteção tradicionais baseados em palavras-

chave e em cartões de identificação.

Os sistemas biométricos têm imenso potencial porque oferecem vantagens em

relação aos sistemas de proteção convencionais, por utilizarem padrões

intrínsecos do utilizador, sendo por isso muito difíceis de duplicar. A

impossibilidade de se poderem perder ou de serem esquecidos são outras grandes

vantagens dos sistemas de reconhecimento que utilizam as características

biométricas.

O sistema desenvolvido pode ser utilizado isoladamente ou combinado com outros

sistemas. Foi desenhado para ter um custo de Hardware muito baixo em relação

aos sistemas que atualmente existem no mercado. Apesar da diferença de custo, os

resultados mostram que é suficientemente fiável para poder ser utilizado em

aplicações reais.

Palavras-chave

Reconhecimento biométrico, Biometria baseada na mão, Veias da palma da mão, Web-

Camera, Geometria da mão.


iv João Ricardo Gonçalves Neves

Table of Contents

1. Introduction ................................................................................................................. 1

1.1. Biometric Systems on the Market ..................................................................... 2

1.1.1. Finger Vein Recognition System ............................................................ 2

1.1.2. Iris Recognition System .......................................................................... 6

1.2. Objectives of the Dissertation ........................................................................... 9

1.3. Contributions of the Dissertation .................................................................... 10

1.4. Structure of the Dissertation ........................................................................... 10

2. Biometric Systems .................................................................................................... 13

2.1. Biometrics Overview ........................................................................................ 13

2.1.1. Social Acceptance and Privacy Issues ................................................. 15

2.1.2. Architecture of a Biometric Recognition System ............................... 16

2.1.3. Performance Evaluation ....................................................................... 17

2.1.4. Comparison between biometric systems ............................................ 19

2.2. Traits used in Hand Recognition Techniques ................................................... 20

2.2.1. Recognition Based on Hand Vein Patterns ......................................... 21

2.2.2. Recognition Based on Hand Palmprints ............................................. 21

2.2.3. Recognition Based on Hand Geometry ................................................ 22

3. State of the Art ........................................................................................................... 23

3.1. Image Acquisition............................................................................................. 23

3.2. Preprocessing ................................................................................................... 27

3.3. Feature Extraction............................................................................................ 31

3.4. Matching .......................................................................................................... 33

4. Proposed Biometric Identification System ............................................................. 37


v João Ricardo Gonçalves Neves

4.1. Potential Applications ...................................................................................... 37

4.2. System Architecture ......................................................................................... 38

4.2.1. Image Acquisition System .................................................................... 39

4.2.2. Preprocessing ........................................................................................ 42

4.2.3. Feature Extraction ................................................................................ 49

4.2.4. Matching System ................................................................................... 53

5. User Interface ............................................................................................................ 55

6. Experimental Results ................................................................................................ 61

6.1. Database Creation ........................................................................................... 61

6.2. Performance Evaluation .................................................................................. 61

6.3. Operating Point Selection ................................................................................ 66

7. Plans for the Future .................................................................................................. 69

8. Conclusions ................................................................................................................ 71

9. References .................................................................................................................. 73


vi João Ricardo Gonçalves Neves

List of Figures

FIGURE 1 - FINGER VEIN SYSTEM DEVELOPED BY HITACHI THAT IS BEING USED IN POLAND [9]. ... 3

FIGURE 2 - HITACHI USB FINGER READER [4]. ..................................................................................... 3

FIGURE 3 - FINGER VEIN CAPTURE METHOD [4]. ................................................................................. 4

FIGURE 4 - ILLUMINATION TECHNIQUE [4]. .......................................................................................... 5

FIGURE 5 - BLOCK DIAGRAM OF FINGER VEIN AUTHENTICATION [4]. ................................................. 5

FIGURE 6 - HUMAN EYE [42]. .................................................................................................................. 7

FIGURE 7 - AOPTIX TECHNOLOGIES INSIGHT® DUO [10] .................................................................. 8

FIGURE 8 - VB I-MATCH DEVELOPED BY VISUAL BOX. [18] ............................................................. 8

FIGURE 9 - DEVELOPED ASSEMBLY. ........................................................................................................ 9

FIGURE 10 - BIOMETRIC SYSTEM MAIN MODULES. ..............................................................................17

FIGURE 11 – EXAMPLE OF FAR AND FRR FOR DIFFERENT THRESHOLD VALUES............................18

FIGURE 12 – EXAMPLE OF ROC CURVE ...............................................................................................19

FIGURE 13 – FAR, FRR AND FTE VALUES FOR DIFFERENT BIOMETRIC TECHNIQUES [4].............20

FIGURE 14 - THE LINES PATTERN OF THE PALMPRINT. 1-HEART LINES, 2-HEAD LINE, AND 3-LIFE

LINE.[19] ........................................................................................................................................22

FIGURE 15 - TYPICAL RECOGNITION SYSTEM ARCHITECTURE ..........................................................23

FIGURE 16 – IMAGE ACQUISITION SETUP FOR THE REFLECTION APPROACH (LEFT) AND

TRANSMISSION APPROACH (RIGHT)[1]. ......................................................................................23

FIGURE 17 - INFRARED PALM IMAGES CAPTURED BY THE REFLECTION (LEFT) AND TRANSMISSION

(RIGHT) METHODS [1]. .................................................................................................................24

FIGURE 18 – MODRIS ET AL EXPERIMENTAL SETUP OF PALM VEIN INFRARED IMAGE ACQUISITION

[1]. ..................................................................................................................................................25

FIGURE 19 - HUAN ZHANG ET AL. HARDWARE SETUP [2]. .................................................................27

FIGURE 20 – OBTAINED ROI AREA WITH MAURICIO RAMALHO APPROACH [17]. .........................28

FIGURE 21 - HUAN ET AL. ACQUIRED IMAGE WITH THE INSCRIBED CIRCLE [2]. ..............................29

FIGURE 22 – NORMALIZED ROI THAT WILL BE USED IN THE FEATURE EXTRACTION MODULE [2].

.........................................................................................................................................................30


vii João Ricardo Gonçalves Neves

FIGURE 23 – (A) THE ORIGINAL PALMPRINT, (B) PALM PRINT AFTER THE CONTRAST ADJUSTMENT

AND SMOOTHING FILTERS [43]. ...................................................................................................30

FIGURE 24 – FEATURE EXTRACTION USING OLOF AT [24]. .............................................................32

FIGURE 25 – FINAL PREPROCESSING STAGES. (A) TARGET IMAGE. (B) BINARIZED IMAGE. (C)

FILTERED IMAGE. (D) NOISE ELIMINATION. (E) THINNED IMAGE. (F) REPAIRED IMAGE [2].

.........................................................................................................................................................33

FIGURE 26 - DEVELOPED SYSTEM ARCHITECTURE ..............................................................................38

FIGURE 27 - DEVELOPED PALM VEIN PATTERN ACQUISITION ASSEMBLY. ........................................39

FIGURE 28 - SYSTEM ILLUMINATION. ...................................................................................................40

FIGURE 29 – SQUARE INFRARED FILTER THAT NEEDS TO BE REMOVED IN ORDER FOR THE

WEBCAM TO CAPTURE INFRARED IMAGES. ..................................................................................41

FIGURE 30 - PHOTOGRAPHIC FILM USED TO FILTER OUT VISIBLE LIGHT. .........................................41

FIGURE 31 - MODIFIED WEB CAMERA USED TO DO HAND VEINS ACQUISITION.................................42

FIGURE 32 - DEVELOPED PREPROCESSING STAGES. ............................................................................42

FIGURE 33 - RAW IMAGE AND IMAGE AFTER ADJUSTMENT STEP. ......................................................43

FIGURE 34 - IMAGE SMOOTHED BY A WIENER FILTER. .......................................................................43

FIGURE 35 - HAND SEGMENTED IN FOREGROUND AND BACKGROUND. .............................................44

FIGURE 36 - HAND CONTOUR. ...............................................................................................................44

FIGURE 37 -ELLIPSE WITH THE SAME NORMALIZED SECOND CENTRAL MOMENT AS THE HAND

REGION. ...........................................................................................................................................45

FIGURE 38 - FIXED POINT MARKED AS THE HALF RED CROSS. ............................................................46

FIGURE 39 - HAND REFERENCE POINTS................................................................................................46

FIGURE 40 - REGION OF INTEREST ACQUISITION. ................................................................................47

FIGURE 41 - ROI TREATMENT STEPS. ..................................................................................................48

FIGURE 42 - REFERENCE POINTS USED TO CALCULATE THE HAND GEOMETRY CHARACTERISTICS

VALUES. ...........................................................................................................................................49

FIGURE 43- OLOF OUTPUT IN THE THREE DIRECTIONS, AND WITH SCALE

RATIO EQUAL TO 3. ........................................................................................................................50

FIGURE 44 - OLOF OUTPUT IN THE THREE DIRECTIONS, Θ=Π/6, Θ=Π/3 AND Θ=0 WITH SCALE

RATIO EQUAL TO 2. ........................................................................................................................50


viii João Ricardo Gonçalves Neves

FIGURE 45 - FAR VALUES FOR DIFFERENT SCALE RATIOS. .................................................................51

FIGURE 46 - FRR VALUES FOR DIFFERENT SCALE RATIOS. .................................................................51

FIGURE 47 - FAR VERSUS FRR FOR DIFFERENT TEMPLATE SIZES. ...................................................52

FIGURE 48 - DATABASE EXAMPLE. ........................................................................................................52

FIGURE 49 - INITIAL INTERFACE MENU ...............................................................................................56

FIGURE 50 - ENROLL INTERFACE ..........................................................................................................56

FIGURE 51 - CORRECT HAND PLACEMENT ..........................................................................................57

FIGURE 52 –THREE POSSIBLE HAND CONTOURS. ................................................................................58

FIGURE 53 - EXAMPLE OF A BAD IMAGE ACQUISITION. .......................................................................58

FIGURE 54 - MATCHING RESULT INTERFACE ......................................................................................59

FIGURE 55 - DATABASE CONTROL INTERFACE AFTER DETECTING AN ERROR. ................................60

FIGURE 56 - DATABASE CONTROL INTERFACE AFTER NOT DETECTING ANY ERROR. ......................60

FIGURE 57 - RECEIVER OPERATION CHARACTERISTIC CURVE FOR A ROI WITH 128X128 PIXELS.

.........................................................................................................................................................63

FIGURE 58 - FAR AND FRR AT DIFFERENT OPERATING THRESHOLDS. ............................................64

FIGURE 59 - ROC CURVE FOR DIFFERENT ROI DIMENSIONS. ............................................................65

FIGURE 60 - FRR (%) AGAINST FAR (%) TO OBTAIN EER FOR DIFFERENT ROI DIMENSIONS. .65

FIGURE 61 - HAND TEXTURE RECOGNITION SYSTEM USING A REGULAR LAPTOP COMPUTER

CAMERA. ..........................................................................................................................................69


ix João Ricardo Gonçalves Neves

List of Tables

TABLE 1 - COMPARISONS BETWEEN THE FRR, FAR AND FTE OF DIFFERENT BIOMETRIC DEVICES

[4]. ..................................................................................................................................................19

TABLE 2 - FAR, FRR AND GAR FOR DIFFERENT THRESHOLD VALUES. ...........................................62

TABLE 3 - VALUES OF FAR AND FRR FOR DIFFERENT OPERATING POINTS. ....................................66


x João Ricardo Gonçalves Neves

List of Acronyms

ATM Automated Teller Machine

CMF Complex Matched Filter

FAR False Accept Rate

FRR False Reject Rate

FTE Failed To Enroll Rate

GAR Genuine Accept Rate

HD Hamming Distance

OLOF Orthogonal Line Ordinal Features

ROC Receiver Operating Characteristic

ROI Region of Interest

XOR Bitwise Exclusive Disjunction


1 João Ricardo Gonçalves Neves

1. Introduction

Nowadays, we can access our personal data from almost everywhere. This is very

convenient but entails increasing risks since the probability of phishing credentials

increases with the number of users. More sophisticated protection systems are

required to control possible harassments, such as ID cards cloning, theft or

compromised passwords.

When thinking about digital protection, one which immediately comes to mind is

the use of passwords and smart cards, since they are used daily for almost

everything. Despite being used very frequently, passwords and smart cards are a

relatively insecure method of protection and access control.

The biometric systems experienced a significant growth in the recent years, both at

research and commercial level, pushed by the need for innovative and improved

ways to protect our personal information.

The word biometric comes from the Greek words ‘bio’ (life) and ‘metric’ (to

measure). The field of biometrics recognition deals with the identification of a

human by using its distinctive traits. They can be categorized in two major groups,

behavioral and physiological.

The behavioral traits are related to the user behavior and include the signature or

gait. The physiological traits include personal characteristics like hand geometry,

fingerprints, ear or face.

The first biometric systems remount to around 29.000 BC, where the primitive

humans used handprints to sign their drawings in cave walls [39]. Much later,

around 500 BC, in Babylonia, fingerprints were used to sign business transactions.

More recently, in the end of the 19th century, Juan Vucetich, an Argentinian police

official, used the prisoners’ fingerprints to catalog Argentina’s criminals [37], and



this technic became a standard identification system which is used all over the

world.

Biometric systems that analyze traits like the finger veins, fingerprint or iris, are

nowadays mostly used as a form of recognition. Those systems are widely used to

control the access to certain applications, private areas or even in forensic

scenarios. The majority of the systems available provide real time automatic

solutions which extract a human feature, then compute a template and compare it

with the ones previously stored in a database to provide a matching decision.

1.1. Biometric Systems on the Market

A wide range of biometric recognition systems are already available in the market

and can be found in the most varied places, like ATM machines or in Airport

passport controls. They all exploit human features to identify an individual and

their reliability is usually provided by expensive equipment.

The following subsections 1.1.1 and 1.1.2 describe three commercial systems. The

first system is unimodal and uses the finger veins as the biometric trait , the second

one is bimodal and uses the features provided by the face and iris and the third one

is also bimodal and uses the face, iris and palmprint as the biometric traits.

1.1.1. Finger Vein Recognition System

One example of a biometric system being commercially explored is the finger vein

recognition system, developed by Hitachi for application in ATM machines [4]. The

finger vein system was developed through Hitachi’s research activities in the area

of medical scanning. In 1997 while the researchers were doing studies about infant

brain activity, they found out that changes in blood flow could be examined using

high intensity near infrared light. It took around eight years of research and

development to create the commercial application. This new recognition system is

already being used in 75% of the bank branches in Japan, making it the market

leading biometric technique in that sector [4]. The users are able to withdraw



money with just a fast scan from one of their fingers, as illustrated in Figure 1,

below. The finger vein system was implemented in order to diminish the

unauthorized Automated Teller Machines (ATM) withdrawals that had increased a

lot in the last years in Japan, due to ATM skimming devices that capture card data

and PIN’s in compromised ATM machines. Nowadays the Hitachi system already

crossed frontiers and it is being used in ATM in other countries, such as Poland and

Turkey [40].

Figure 1 - Finger vein system developed by Hitachi that is being used in Poland [8].

The Hitachi finger vein reader is also being used in other applications like door

openings or even to login into computers using a USB device, as shown in Figure 2.

Figure 2 - Hitachi USB finger reader [4].

By using a biometric trait like the finger vein pattern, the Hitachi system provides a

great help in the protection against fraudulent approaches to access the private

data of a legitimate user. The used trait has the advantage of being internal to the

human body and invisible except under very specific conditions. This system also



requires the presence of live blood vessels, decreasing the remote chance of a

possible forgery.

The accuracy of a system used in a delicate area such as money withdrawal is a

very sensitive matter. The vein based system can meet all the accuracy

requirements because, with the right illumination conditions, there is an

extraordinary degree of variation between patterns which reinforces their

distinctiveness. In addition, the finger vein patterns do not change through all the

adult life time.

Another big advantage is the insensibility of this system to external factors like

dirt, sweat or grease of the finger. It is even possible to use this system using latex

gloves which increases the hygienic component of the system.

The Hitachi system works by illuminating the finger with near infrared light as can

be seen in Figure 3. The output of this system is an image with a distinctive pattern

that will be used to do the matching of the users.

Figure 3 - Finger Vein Capture Method [4].

Hitachi researchers found out that the best images are obtained by shining light

through the finger as illustrated in Figure 4. Hitachi developed a side illumination

system to address the problem of putting the finger inside the device. This

technique still uses the advantages of using transmitted light with the complement

of having an open and suitable device.



Figure 4 - Illumination Technique [4].

To make the finger device usable for all kind of finger sizes and environments, the

light source intensity is adjusted automatically. This adjustment provides the

optimization of the image contrast, a higher image detail and the minimization of

the noise.

The authentication process of Hitachi was developed in four main steps: the

capture of the finger vein image, the normalization of the image, the feature

extraction and the matching, as depicted in Figure 5. In this case, the reference

template is stored in the smartcard to increase the level of protection.

Figure 5 - Block diagram of finger vein authentication [4].

Finger vein images are captured and moved into the CPU memory. After having the

images in memory, the algorithm dynamically adjusts the brightness of the

illumination source to improve the quality of the acquired image.



In the second stage the finger vein image is normalized to accommodate all the

possible geometric changes in position or angle of the finger. The normalization is

achieved by identifying the outline of the finger in the acquired image, and then

rotating the whole image to normalize the slope of the outline.

The distinctive features of the finger blood vessel pattern, which will be used in the

matching stage, are extracted in the third stage. This step is crucial to eliminate the

variations provided by changes in the body metabolism or by the image conditions.

The result of the extraction step is a standard finger vein template of nearly 400

bytes which is appropriate to be used in the matching algorithm.

In the fourth and last stage the captured finger vein template is matched against

templates, stored in the database. If the matching score is below the predefined

threshold the user is successfully authenticated.

If the user is using a smartcard, the matching verification uses the template stored

inside the card. This provides extra security, since the reference template never

leaves the card, but on the other hand it reduces the security in the case of a

possible cloning of the card. Alternatively, the reference templates can be stored in

the finger vein device itself, on an attached PC, or somewhere else on the network.

All these storage approaches are vulnerable to a possible database breach.

1.1.2. Iris Recognition System

There are several companies that are developing and selling systems that rely on

the human iris to do people recognition. The iris is a thin circular diaphragm,

which lie between the cornea and the lens of the human eye. The iris is perforated

close to its center by a circular aperture known as pupil [41]. The human eye

description is depicted in the Figure 6.



Figure 6 - Human eye [41].

The AOptix Technologies Insight Duo according to AOptix [10] is the first biometric

system that captures simultaneously iris and face images, adding a recognition

quality of a standard-based face record to the unparalleled uniqueness of iris

recognition.

The InSight Duo depicted in the Figure 7, was created in order to be used in all

kinds of environments, even with non-technical or non-acclimated users. This

system is currently used in the Gatwick airport in the south of London with the

goal of improving the overall airport experience.



Figure 7 - AOptix Technologies InSight® Duo [9]

Another company, VisionBox, developed a hardware system called VB I-MATCH

[17], illustrated in Figure 8, that in addition to iris recognition also supports

fingerprints and facial recognition. This system is already being used at Schiphol

airport and also in land, air and sea borders in Portugal, UK, Finland and Norway

among others.

Figure 8 - VB I-MATCH developed by Visual Box. [17]



1.2. Objectives of the Dissertation

This dissertation proposes to develop a low cost hand vein biometric recognition

system intended to do a highly efficient identification and to have a fast response

and easy usage. The application is native to Matlab and was compiled to be used in

a Windows operating system.

In this system the hand needs to be placed inside a special assembly, developed in

this thesis, which is shown in Figure 9. Inside the assembly the hand is exposed to

near infrared light, provided by high power leds. The assembly is essential to

control the illumination intensity, vital to acquire quality images.

Using the vein patterns as the biometric trait has a lot of advantages over the most

commonly used fingerprint and palmprint verification systems. Those advantages

are:

The veins are invisible except under special circumstances.

The system requires live blood vessels to work.

Despite being a low cost system the results obtained proved that the system is

reliable and it has a great potential. The target applications for this system are

countless. Opening doors, allowing ATM money withdrawals or unlocking a

computer are just some examples.

Figure 9 - Developed assembly.



1.3. Contributions of the Dissertation

The main outcomes of this dissertation are:

An accurate low cost recognition system that uses the hand vein pattern as

the biometric trait.

An assembled prototype designed to acquire the hand vein pattern in

controlled conditions.

A Windows application with a user friendly interface.

A proof of concept based on results obtained with the system.

1.4. Structure of the Dissertation

This dissertation is organized in the following sections:

Chapter 2 - Biometric Systems: A brief review of the biometric

recognition systems available, the related problems, requirements and the

typical architecture.

Chapter 3 - State of the Art: A review of different techniques implemented

in the literature that might be used while implementing a biometric

recognition system.

Chapter 4 - Proposed Biometric Identification System: Description of

the architecture and implementation of the developed biometric

recognition system.

Chapter 5 - User Interface: In this section the user interface of the

developed system are revealed and explained.

Chapter 6 - Experimental Results: The experimental results obtained with

the developed system are presented and compared with the previous work

from the literature.



Chapter 7 - Plans for the Future: Possible upgrades and improvements for

the developed system are proposed in this chapter.

Chapter 8 - Conclusions: Conclusions about the developed work are

drawn.



2. Biometric Systems

Biometric systems are emerging as a new form of identification and access control,

already being used in a wide range of applications.

In this chapter the importance of biometric systems is analyzed, as well as their

requirements and challenges. The typical biometric system architecture is also

described, as well as the most commonly used performance evaluation techniques.

2.1. Biometrics Overview

Passwords and smart cards are nowadays the default ways of authentication, used

to grant access to protected information. Passwords are used commonly because

they are a simple and inexpensive mechanism to implement and use. However,

mostly due to weak characters combinations and poor password practices, they

are known for being a poor protection method [11]. They are also easy to steal and

forget. Biometric recognition systems surpass these problems because they only

depend on biological and behavioral characteristics, which are inherent to the

human individuals.

Any human characteristic can be used as a biometric characteristic as long as it

fulfills the following requirements [12]:

Universality - Every user should have the required trait.

Uniqueness - The trait used should be sufficiently different for each user.

Permanence - The trait used must be reasonably invariant over time.

Measurability - The trait must have an easy acquisition and measurement

and in addition, the obtained data must be easy to process.

Performance - The system must be accurate, fast and robust.



Acceptability - The users must feel that the technology is useful and secure

in order to be compelled to have their biometric trait captured and

assessed.

There are a lot of human characteristics [12] that might be used in a biometric

system to identify an individual. These characteristics might be used on their own

(unimodal system) or in addition to each other to provide a stronger system

(multimodal system). Examples are:

Hand Palm Veins – Uses the vascular system patterns of the hand palm for

recognition.

Fingerprint – Uses the ridges and valleys pattern from the surface of the

fingertip.

Palmprint – Works in the same way as the fingerprints but uses a larger

area in the user’s palm.

Hand geometry – Works using a number of measurements taken from the

human hand, like its shape, size, length and widths of the fingers.

Iris Recognition – Uses the iris unique patterns to do the matching.

Face Characteristics – Utilizes the facial features for the recognition.

Gait – Exploits the movement characteristics of the user while walking.

DNA – Uses the diversity of the DNA characteristics.

Signature –Exploits the differences between users hand writing.

Voice – Uses the differences between the users acoustic spectrum of the

voice.

Depending on the goal for the given biometric recognition system, it may operate

in verification or identification mode.

Identification Mode – The system attempts to recognize an unknown

person by searching all the templates in the database for a possible match.



Verification Mode – The system validates a user identity by comparing his

captured biometric data with a stored template. Instead of searching the

entire database for a template match, like in the identification mode, the

system directly accesses the template associated to a username or identity

card.

The biometric system needs to acquire data from the user. While capturing the

data, the interaction between the user and the system may create problems of

social acceptance or privacy issues to the users, as discussed in the next

subsection.

The typical biometric system architecture, some performance evaluation

methodologies and some hand recognition approaches are also discussed in the

following subsections.

2.1.1. Social Acceptance and Privacy Issues

The user willingness to use a biometric system is related to the interface easiness

and the comfort of the acquisition procedures. The systems that do not need

contact, like those using voice or iris images, are the most accepted because they

are more hygienic and user-friendly. On the other hand, systems that capture the

user characteristics without his perception are perceived as a threat to the privacy

by many users.

The privacy issue must be taken very seriously because the characteristics

obtained through biometric recognition systems may be used to provide additional

information about the individual. One good example of this problem is the retinal

pattern that may provide medical information about diseases of the user (e.g.

diabetes or high blood pressure). This is the kind of information that a health

insurance company could use in an unethical way to deny some benefits to an

individual that has a great risk of becoming sick [38].



To prevent user’s fears of having the biometric identifiers compromised and used

in different systems and databases, the majority of the companies on the market

using this kind of technology do not store the physical characteristics in the

original form, but instead they store a template in an encrypted format, in order to

make the recreation of the original characteristics impossible.

2.1.2. Architecture of a Biometric Recognition System

Usually, a standard biometric recognition system is composed of six main modules,

according to the generic architecture illustrated in Figure 10:

Data Collection – To use a biometric system, it is essential to capture the

biometric data from a biometric sensor. In this step it is necessary to

correct errors, related to human factors, environmental conditions or even

due to the quality of the sensor used.

Preprocessing – This module is where tasks like image alignment,

enhancement or region of interest identification take place.

Feature extraction – The feature extraction module is where the region of

interest is extracted and converted on a suitable template.

Storage of the data in the database – In this module the biometric data is

stored in the database. A vector of numbers or an image with particular

properties is used to create the template, which is a synthesis of the

relevant characteristics extracted from the source. Elements of the

biometric measurement that are not used in the comparison algorithm are

discarded from the template in order to reduce the file size and to protect

the identity of the enrollee.

Matching – In this module the system checks the database for similar

templates. The matching is processed by computing a similarity score

between the new and the stored templates.



Decision – The system verifies if the similarity score is greater than the

predefined threshold (t). If the score is lower than t, the template obtained

and the template stored are assumed to belong to the same individual.

Figure 10 - Biometric system main modules.

2.1.3. Performance Evaluation

Different measurements of the same individual, taken at different times will never

be exactly identical. To overcome this problem, a similarity score between the two

measurements is calculated. If that score is above a predefined threshold t it is

assumed that the two measurements do not belong to the same person. If the

score is below the threshold it is the other way around.

The False Accept Rate (FAR) and the False Reject Rate (FRR) can be used to

measure the accuracy of a biometric recognition system. FAR and FRR are both

functions of the system threshold t:

FAR - Is the probability of a successful access attempt by an impostor. The

impostor access happens if the similarity score between his template and a

genuine user’s template is less than the threshold.

FRR - Is the probability of a failed access attempt by a genuine user. An

incorrect reject happens when the score between the actual template and



the template of the same user stored in the database is larger than the

threshold.

If the threshold value is decreased, making the system less tolerant to input

variations the FAR decreases, making the system more secure but, on other hand,

the FRR increases creating more difficulties to genuine users. The rate at which the

FRR equals the FAR is called the Equal Error Rate (EER). The FAR and FRR values

for different thresholds is depicted in the Figure 11.

The Failed to Enroll rate (FTE) is also an important measure to test the system

performance. It measures the percentage of unsuccessful attempts during the

creation of a template from a recently acquired image.

Figure 11 – Example of FAR and FRR for different threshold values.

The system’s response at all thresholds can be represented by a Receiver

Operating Characteristic (ROC) curve. That curve is a plot of the Genuine Accept

Rate (GAR) that is equal to (1-FRR) versus FAR for various threshold values. A ROC

curve is depicted at Figure 12. A perfect system would use an operating point

where the GAR would be 100% and the FAR 0%.

0

10

20

30

40

50

60

70

80

90

100

0 10 20 30 40 50 60 70 80 90 100

%

Theshold (%)

FAR

FRR



Figure 12 – Example of ROC Curve

2.1.4. Comparison between biometric systems

Table 1, provided by the International Biometric Group [3], presents a comparison

among some of the most used commercial biometric systems, highlighting their

differences.

Table 1 - Comparisons between the FRR, FAR and FTE of different biometric devices [4].

Table 1 shows that FAR is typically in the order of 0.01% and the FRR is below 2%,

showing that the systems are very resistant to impostors, although that can create

some trouble to genuine users. Figure 13 shows the same information graphically.

0

20

40

60

80

100

0 20 40 60 80 100

GA

R(%

)

FAR(%)

ROC

ROC



Figure 13 – FAR, FRR and FTE values for different biometric techniques [4].

The results presented by IBG [3] (see Table 1 and Figure 13) were obtained in

simulation scenarios, similar to real situations.

The best techniques, according to [3] are clearly the finger vein and the palm vein

due to the lowest FAR, FRR and FTE values. The finger vein got better FRR results

than the palm vein because it is easier to control the illumination on a smaller area.

The results presented above clearly show that the palm vein approach selected for

this thesis is a good choice.

2.2. Traits used in Hand Recognition Techniques

Palm vein patterns, palmprint or even the hand geometry, can be used to identify

an individual. If those characteristics are used wisely, they might be combined to

become the input of a multi-biometric technique and be used for identification

purposes. Those techniques are discussed with more detail in the following sub-

sections.



2.2.1. Recognition Based on Hand Vein Patterns

A hand vein recognition system uses the vascular patterns of an individual as

personal identification data. The palm veins are convenient because they have a

complex vascular pattern and thus have a lot of unique features that can be used

for personal identification. Their falsification is also very difficult.

The hand vein detection process is based on a camera that takes a picture of the

subject’s veins under a source of infrared radiation at a specific wavelength. This

system exploits the vascular system to work. In the human physiology the

hemoglobin present in the blood is oxygenated at the lungs and then conducted to

the tissues of the body through the arteries. After the oxygen is released to the

tissues, the deoxidized hemoglobin returns back to the heart through the veins.

The rate of absorbency is different in the two types of hemoglobin. Deoxidized

hemoglobin absorbs light at a wavelength of about 760nm in the near infrared

region which is crucial for the system to work properly. The palm vein system is

able to detect veins but not arteries due to the specific absorption of infrared

radiation in blood vessels. We can use this technique to almost every part of the

body, however the hand is the most suitable body part because it is generally

available [2].

2.2.2. Recognition Based on Hand Palmprints

A hand palmprint recognition system uses the palmprint features to uniquely

identify an individual. Principal lines, wrinkles and ridges are shown in the Figure

14. The three principal lines of the palm are called the heart line, the head line and

the life line. These lines are unique and hard to miss and they almost don’t change

through the whole life of a person, which makes them a really good tool to build a

biometric identification system. The wrinkles are thinner than the principal lines

and are more irregular. Besides wrinkles and principal lines there are the ridges

that exist all over the palm.



Figure 14 - The lines pattern of the palmprint. 1-heart lines, 2-head line, and 3-life line.[18]

Hand recognition can be based on the palmprint statistical features or on

palmprint structural features [18].

2.2.3. Recognition Based on Hand Geometry

Hand geometry recognition systems are based on measurements taken from the

human hand, including its shape, size of palm and length and widths of the fingers.

This technique has the advantage of being very simple and inexpensive.

Environmental factors such as dry weather or dry skin do not affect the

performance of the verification accuracy. This kind of system alone is not very

strong, but in addition to another system like the hand veins recognition system

can be very useful. This technique can work like a filter, or a soft-biometric,

excluding the users that have hands with a very different geometry from the one

recently acquired [19].



3. State of the Art

The typical high-level architecture of most hand veins biometric systems is the one

depicted in Figure 15, below. It includes modules for image acquisition,

preprocessing, feature extraction and for the matching stage. Most differences

between biometric systems lay on the different approaches taken for each of the

blocks. In this chapter different techniques to process each of the different

modules are discussed.

Image Acquisition PreprocessingValid Image

Yes/No

Feature Extraction

Templates Database

User FoundYes/No

Matching

Access Granted

Access Denied

Figure 15 - Typical Recognition System Architecture

3.1. Image Acquisition

Infrared images of the veins can be obtained through the light reflection or

transmission methods.

In the reflection approach a light source and the camera are placed at the front of

the target, while in the transmission case it is located at the back of the target.

Figure 16 illustrates both cases.

Figure 16 – Image acquisition setup for the reflection approach (left) and transmission approach

(right)[1].



According to Modris et al. [1], the transmission method allows obtaining better

results because it shows deep veins, while the reflection systems only shows

surface veins. One of the disadvantages of the transmission method, however, is

the necessity of a higher power light source, which may consume considerably

more power than the reflection method.

The system dimension is another disadvantage of the transmission technique. The

reflection method can be implemented in smaller spaces, because all the required

components can be attached on the camera side. Otherwise, the transmission

method needs to have the light source behind the palm which will increase the

total size of the system. Only the reflection method provides the capacity of

developing biometric systems for small devices, like mobile phones.

The amount of data that can be collected from a single image might also be a

problem of the transmission technique due to the acquisition of the bone structure,

which might hide the vein pattern that is used in the recognition process. The

captured images of the two techniques are depicted in the Figure 17.

Figure 17 - Infrared palm images captured by the reflection (left) and transmission (right) methods

[1].

Modris et al. [1], after taking into consideration the advantages and the drawbacks

of the transmission and reflection approaches, selected the reflection method over

the transmission method because of the power consumption and the size of the

system.



The hardware setup proposed by Modris et al. [1] displayed in the Figure 18,

consists of a CCD camera, that needs to be sensitive in the near infrared spectrum,

an IR lens, an illumination system composed by IR LEDs, an IR band pass filter of

850nm wavelength and a palm fixing stand that is used to simplify the image

recognition task by avoiding the preprocessing tasks related to the rotation and

translation correction.

Figure 18 – Modris et al. experimental setup of palm vein infrared image acquisition [1].

After acquiring the images they are transferred to a PC to be saved on a database

and to be used in further processing. The image resolution used before selecting

the region of interest is 640 x 480 pixels.

According to Huan Zhang et al. [2] the hardware setup plays a very important part

in the design of a palm vein recognition system. The components of the image

acquisition module are a near-infrared camera and the illumination system. The

most important attribute of the camera used to acquire the images must be the

response to near infrared radiation. Attributes like spatial resolution and frame

rate are less important because the image must be still and the vein pattern details

are detected at short distance even with low resolution cameras.



The arrangement of the illumination system in Huan et al. approach [2] has also a

great importance in the image acquisition process, since it will provide accurate

contrast between the veins and the surrounding tissue while keeping the

illumination errors at a minimum.

In order to avoid the influence of visible light that might compromise the quality of

the acquired image, the camera must have a low response in the wavelength of the

visible light. In [2], the authors use a JAI AD-080 CL 1/3’ CDD near-infrared

camera to avoid the described problem.

Due to the optical properties of the human skin, near-infrared light cannot

penetrate very deep in the human tissues, making the extraction of deep vein

patterns a very difficult task. The patterns used in [2] are mainly from superficial

veins. Due to the illumination setup used in [2] the statistical maximum distance of

penetration obtained is 3 mm, which will be a limitation of the quantity and quality

of the extracted blood vessels pattern.

The optical absorption and scattering coefficients must be taken into account while

doing the vein feature acquisition. The first one determines how far light can travel

under the human skin before losing its intensity. The second coefficient

determines how far light can travel before losing its original phase and had a

change in its direction.

The optical properties described above imply that the illumination conditions must

be homogeneous through the whole region of interest area and must be similar

through different acquisitions. The contrast must be high enough to provide a

reduction of the complexity of the preprocessing algorithms.

Huan Zhang et al. [2] developed a system that uses 850 nm near-infra LEDS as the

light source. To avoid the problems of the non-homogeneous illumination provided

by the LEDS, a holographic diffuser is used to provide constant illumination. The



holographic diffuser scatters the light from the leds and diminishes the radiation

intensity.

In order to increase the uniformity of the acquired image, all the equipment is

inside a closed box (see Figure 19), painted with a black paint with high absorption

rate. The user hand is placed inside a small rectangular opening in front of the

camera.

Figure 19 - Huan Zhang et al. hardware setup [2].

3.2. Preprocessing

In order to use the captured palm images for a biometric application, the palm vein

pattern has to be extracted and segmented. This process is not trivial since the

blood vessels are almost undistinguished in some images. This section will review

some techniques that can be used in order to preprocess the acquired images

before the feature extraction module.

In Mauricio Ramalho’s proposed system [16] the preprocessing stage starts with

an image adjustment, where the image is converted to gray scale, resized and

filtered. The acquired image is resized to a maximum of 256x256 pixels in each

direction, in order to reduce the computational effort required. After resizing the

image, a low-pass Wiener filter is applied in order to eliminate the noisy areas,

smooth the textures and highlight the contrast.



After the initial steps, the image is segmented into background and foreground.

Through a constant black background the hand region is acquired through an

automatic global histogram thresholding technique. This is accomplished by using

the Otsu’s thresholding method [43], which chooses the threshold value that

minimizes the intra-class variance of the output binary image.

The binary image obtained is the input of an algorithm based on morphological

reconstruction [21] that fills the holes, which might be present in the foreground

image area. After the reconstruction, the major object in the image (hand) is

selected and the hand contour is obtained.

After identifying the hand in the image, the region of interest (ROI) must also be

identified. The reference points that will be used in the ROI identification (Figure

20) will be acquired through two different techniques: The radial distance to a

fixed point [14] [15] and the contour curvegram [14].

Figure 20 – Obtained ROI area with Mauricio Ramalho approach [16].

After being detected, the ROI (area inside the blue square in the Figure 20) is then

normalized for comparison purposes. The ROI got to be normalized because



different hands will have different ROI sizes and orientations. The normalization

consists on the rotation of the ROI to a vertical position followed by a resizing to

reduce the computational power required.

Huan Zhang et al. [2] image acquisition (see Figure 21) follows the same basic

preprocessing steps as Mauricio Ramalho [16], but instead of using a Wiener filter,

a Gaussian smoothing filter is used. Despite being different, the objective behind

the use of both filters is the same.

The ROI identification in Huan et al. approach is obtained through a technique

called inscribed circle-based segmentation. This technique calculates the circle that

meets the border of the palm in order to extract the larger area possible. One

advantage of this technique is that the dimension of the radius of the obtained

circle is different from person to person. That information can be used to conclude

almost instantly if two palms belong to the same person.

Figure 21 - Huan et al. acquired image with the inscribed circle [2].

The normalized ROI used in the feature extraction step is a predefined rectangular

area inside the obtained circle, see Figure 22.



Figure 22 – Normalized ROI that will be used in the feature extraction module [2].

Michael Ong et al. approach [42] starts by identifying the ROI. The ROI localization

starts with the segmentation of the hand from the background through a skin-color

thresholding method. After the segmentation, a valley detection algorithm is used

to find the valleys of the fingers. Those points will be the reference points, used for

the ROI detection.

The enhancement of the contrast and the sharpness of the ROI images are obtained

through a Laplacian isotropic derivative operator, followed by the use of a

Gaussian low-pass filter used to smooth the palmprint images and bridge some

small gaps in the lines. Both the original and the preprocessed image are shown in

Figure 23.

Figure 23 – (a) The original palmprint, (b) palm print after the contrast adjustment and smoothing filters [42].



3.3. Feature Extraction

In the feature extraction module the acquired biometric data is processed to

extract a set of discriminatory features for further recognition and matching. This

section will review some techniques that can be applied in order to extract the

necessary data from the acquired images.

There are three main categories of feature extraction techniques: (i) Appearance-

based (or Subspace-based), (ii) Texture-based and (iii) Line-based [27].

The appearance-based approach uses the palm print image as a whole. The most

used methods within this approach are the principal component analysis (PCA)

[34], linear discriminant analysis (LDA) [28] and independent component analysis

(ICA) [35].

The texture-based approach treats the palm print as a texture image. Therefore

statistical methods like Law’s convolutions masks, Gabor filters and Fourier

Transforms could be used to compute the texture energy of the palm print. Ordinal

measure [36] is another powerful method to extract the texture feature. It detects

elongated and line like image regions which are orthogonal in orientation. The

extracted feature is known as ordinal feature.

The line-based approach used by some researchers [29] uses the structural

information of the palm print. The features used are the line patterns, like principle

lines, wrinkles ridges and creases. Other researchers use more flexible approaches

to extract the palm lines by using edge detection methods like Sobel operator [30],

morphological operator [31], edge map [32] and modified radon transform [33].

Mauricio Ramalho [16] followed an appearance-based approach. The normalized

ROI obtained in the preprocessing stage, is converted into a binary vector of

luminance values, which is used as the input for PCA and LDA. This algorithm

linearly transforms the vector into a more discriminating feature space and



reduces the data dimensionality. The set of linearly transformed features are the

templates that will be used in the matching stage.

Nuno Moço [23] used a texture-based approach. The feature extraction and

template creation modules are based on a technique called Orthogonal Line

Ordinal Features (OLOF), originally proposed in [5].

The idea behind this method is to qualitatively compare two elongated line like

image regions, which are orthogonal in orientation and generate one bit feature

code according to the observed differences. The set of code bits will be the

template used in the matching stage. The feature extraction steps using the OLOF

technique are present in Figure 24.

Figure 24 – Feature extraction using OLOF at [23].

Wu et al. [7] use a line-based approach which extracts the palm lines through an

edge detection method. The method used is the Canny edge operator [6] to detect

the palm lines.

Huan Zhang and Dewen [2] also used a line-based approach. After the detection of

the ROI, a Niblack [22] algorithm is used to convert to binary the acquired ROI.

Niblack’s algorithm is a local thresholding method based on the calculation of the

local mean and of the local standard deviation. The binary image is filtered by a

median filter to reduce the noise. The last steps are the region growth used to

remove the regions which are beyond the predefined vein width range and a

thinning method used to thin and repair the vein line. All the steps are depicted in

Figure 25.



Figure 25 – Final preprocessing stages. (a) Target image. (b) Binarized image. (c) Filtered image. (d) Noise elimination. (e) Thinned image. (f) Repaired image [2].

3.4. Matching

The Matching is the last step in a biometric recognition system, where the recently

acquired template is compared with the ones stored in the database to provide a

matching score.

The matching score is a value that will quantify the similarity between the new

template and the templates stored in the database. The value of the matching score

indicates the chance of the two templates coming from the same individual. The

decision is based on a threshold. If the matching score is higher than the threshold,

the new template and the one stored in the database probably do not belong to the

same person.

Depending on the types of the features extracted, a variety of matching techniques

are used to compare two palm print images. In general these techniques can be

divided into two main categories, the Geometry-based matching and the Feature-

based matching.



The Geometry-based matching techniques sought to compare the geometrical

primitives like points [24] and line features [25] on the palms. The point detector

method [26] uses a distance metric such as Hausdorff distance to calculate the

differences between two templates. On the other hand, the line based features

generally use Euclidean distances to compute the similarity between two line

segments. The Line Based matching is perceived as more informative than point-

based matching because the palm print pattern could be better characterized using

the rich line features, compared to isolated datum points [27].

The Feature-based matching works well for the appearance based and texture

based approaches. Researchers which studied the subspace methods like PCA,

LDA, and ICA use mostly Euclidean distances to compute the matching scores [28].

On a successful match it is highly expected that the value of Euclidean distance

should be zero or as low as possible. A smaller value of Euclidean distance

indicates a closest match and a larger value points to a very low probability of

finding a corresponding match.

The Feature-based matching has a great advantage over Geometry-based matching

when low-resolution images are used. This advantage comes from the fact that

Geometry-based matching usually requires higher resolution images to acquire

precise locations and orientations of the geometrical features [27].

When the palm print features are transformed into a binary bit string for

representation, the Hamming distance is utilized to count the bit differences

between two strings. The Hamming distance between two vectors is the number of

differences among the coefficients of the two vectors. If two vectors are equal, the

Hamming distance would be zero. The Hamming distance value is calculated with a

XOR operand between the two vectors [23]. The Hamming distance is obtained

through the equation (1).



∑∑ ( ) ( )

( )

A successful matching is obtained if the Hamming distance result is below the pre-

defined threshold.



4. Proposed Biometric Identification System

This section discusses possible applications for the developed system and

describes the technical options, the architecture and the procedures adopted for

the system development.

4.1. Potential Applications

The developed system is intended to work as an access control for certain areas or

applications. Despite being implemented with a low budget, it can achieve a good

performance. Some possible applications are:

Open Doors – Instead of using the old-fashioned key or even a magnetic card,

which might be lost, stolen or even replicated, the user just have to insert his hand

in the palm vein assembly in order to send a signal to an electronic locket. After

receiving the signal the electronic locket should open the door.

Online Bank Accounts Management – As an alternative of using regular

passwords and matrix cards, which are currently used in most bank corporations

to manage online accounts, the palm veins system could be used as a computer

peripheral to do a secure login.

ATM Machine Operations – This kind of functionality is already being used in

some countries, like Japan, where the system is attached in the assemblies of the

ATM stations to be used as supplement or substitute of the old PIN, which will

theoretically reduce the amount of unauthorized money withdrawals.

Access to High Secured Folders – Despite of being logged in your account, an

extra protection might be put in use to give access to highly valuable

documentation stored in the computer. That extra protection would be given

through the palm vein system that, once again, could be used as a peripheral.



Unlock Computer – Instead of choosing the user name and password to do the

computer login, a brief hand palm vein scan should be sufficient to unlock the

computer.

4.2. System Architecture

The proposed biometric recognition system is unimodal and uses the hand vein

pattern as the biometric trait. The architecture of the developed system is

presented in Figure 26.

Pre-processingImage

Acquisition

Access Granted

Valid Yes/No

Feature Extraction

MatchingValid

Yes/No

Access DeniedBinary

Templates Database

Register

Register Yes/No

Figure 26 - Developed system architecture

The approaches taken for every module of the developed system will be explained

in detail in the following subsections. The following paragraph gives a summarized

description of the approaches taken.

To do the image acquisition in the developed system a modified low cost camera is

used. After the image acquisition, the captured image is resized in order to reduce

the required computational power, turning the preprocessing less demanding and

consequently saving processing time. After resizing the acquired image, it is

preprocessed in order to reduce the amount of noise. The detection of the region of

interest is obtained through some reference points in the hand contour. The



feature extraction and template creation sections are based on the OLOF [5]

technique. As shown in [5], the OLOF method turns the veins representation

robust against illumination variations. It also makes the matching stage effortless

since the dissimilarities between two palmprints can be measured through the

differences in the binary bits from the two templates with a simple XOR operator,

which can be computed almost instantly.

4.2.1. Image Acquisition System

In this work the transmission illumination method was selected over the reflection

method because it shows deeper veins and allows the use of low quality cameras,

although this method requires higher energy consumption and more space.

The image acquisition module developed for this dissertation uses a low cost

webcam (Logitech QuickCam Pro 9000) that is installed in a special assembly (see

Figure 27), in order to operate in controlled illumination conditions.

Figure 27 - Developed palm vein pattern acquisition assembly.

Since the system performs recognition based on vein images, the illumination is

obtained using 15 near IR leds (OSRAM – SFH4550) [20], Figure 28.



Figure 28 - System Illumination.

The box is a cube with 26 cm side. The top of the box was painted black in order to

reduce the interference caused by the visible light coming from the exterior of the

assembly.

In order to be able to capture the near infrared light, necessary for the vein

acquisition, the low cost web-camera requires the removal of the infrared filter

that is placed behind the lens, as illustrated in Figure 29. The main problem

associated with the removal of the IR filter is that the auto-focus functionality of

the web-camera becomes damaged, which turns the capture of good quality images

at long distances impossible. This problem will not affect the image acquisitions of

the developed system since they are captured from a small distance.



Figure 29 – Square Infrared filter that needs to be removed in order for the webcam to capture infrared images.

As the camera needs to detect only infrared light, a visible filter has been applied.

An old fashioned photographic revealed film was used for this purpose, as shown

in Figure 30.

Figure 30 - Photographic film used to filter out visible light.

After removing the IR filter in the back of the lens and assembling the visible light

filter in front of it, the camera is ready to do the acquisition of near infrared

images, Figure 31.



Figure 31 - Modified web camera used to do hand veins acquisition

With this assembly, images are acquired with a resolution of 240x320 pixels.

4.2.2. Preprocessing

The preprocessing stage prepares the image for the feature extraction phase. This

is obtained through several stages: image adjustment, filtering, segmentation,

contour detection, key point’s detection and region of interest extraction, as

represented in the architecture illustrated in Figure 32.

Image Adjustment

Image FilteringImage

SegmentationRaw Image

ROI Acquisition

Valid?Key Points Discovery

Contour Detection

Region of Interest

Extraction

No

Yes

Figure 32 - Developed preprocessing stages.



The first step of the preprocessing stage is image adjustment. During this step the

raw image is resized from 240x320 to 192X256, in order to reduce the

computational power required through the process. After resizing the raw image,

the color space is converted from rgb to grayscale since the luminance information

is enough for the image segmentation, see Figure 33.

Figure 33 - Raw image and image after adjustment step.

The second step of the preprocessing chain is the filtering, used to reduce the noise

of the image and to smooth the areas with little variance. This is obtained using a

Wiener filter (the same type of filter used by Mauricio Ramalho in [16]). The

output of the Wiener filter is depicted in Figure 34.

Figure 34 - Image smoothed by a Wiener filter.

The third module performs image segmentation, where the image is segmented

into foreground and background through a pre-defined threshold. Thresholding is



a very fast way of identifying the hand using the contrast with the black

background. After thresholding the image it is converted to binary. One example of

a segmented image obtained is depicted in Figure 35.

Figure 35 - Image segmented in foreground and background.

The segmented image is the input of the contour detection algorithm [13]. This

algorithm choses a random starting point in the hand boundary and then searches

for all the boundary pixels. The contour is essential for identifying the region of

interest and the reference points. The hand contour can be seen in the Figure 36.

Figure 36 - Hand contour.

The key point’s acquisitions are obtained through the combination of two different

techniques, the radial distance to a fixed point [14] [15] and the contour



curvegram [14]. Both methods identify the fingertips and the valleys between the

fingers.

The radial distance to a fixed point technique calculates the Euclidean distance

between every contour pixel and a fixed point, which is the middle point of the

region where the wrist crosses the edge of the image.

The contour curvegram analyzes the intensity of the curvature along the contour,

and can be constructed by using a technique called difference-of-slopes [14].

The two methods have their benefits and drawbacks, but together they create a

stronger set of reference points. The radial distance to a fixed point is the first

technique used in order to get an approximation of the final reference points. After

obtaining the raw key points, the contour curvegram is used around the obtained

locations. The final obtained positions are the final fingertip and finger-valley

locations.

In order to obtain a good location of the fixed point, to be used in the radial

distance method, an ellipse (Figure 37) with the same normalized second central

moment as the hand region is drawn. Through the hand contour input, the ellipse’s

parameters like the major and minor axes, center position, end-points and lengths,

orientation (given by the angle between the major and minor axes) are calculated.

Figure 37 -Ellipse with the same normalized second central moment as the hand region.



After obtaining the parameters that define the ellipse, it is necessary to find out in

which side of the minor axis the wrist is located. This verification is obtained

through the counting of the contour points that lie on each side of the axis. The

wrist is located on the side with fewer points. Knowing the axis’ side on which the

wrist lies, the fixed point (Figure 38) in the wrist is defined as the intersection

point between the major axis and the edge of the image.

Figure 38 - Fixed point marked as the half red cross.

The additional reference points, represented as yellow dots in Figure 39, are

necessary to extract the palm’s region of interest. These additional reference

points are determined by discovering, the thumb, index and pinkie fingers.

Figure 39 - Hand reference points.



The final set of hand reference points, is composed by the five fingertips, the four

finger valley and the three additional reference points.

After finding the reference points, the square that represents the ROI is obtained

(Figure 40). The square position is defined through a line segment that is drawn

between the index and the pinkie finger.

Figure 40 - Region of interest acquisition.

Different hands will create squares with different sizes and orientations that will

need to be normalized for matching purposes. In order to do the standardization

the ROI is rotated to a vertical position and resized to a standard dimension. The

standard ROI dimension chosen is 128x128, due to the results that will be

presented in the performance evaluation section. Decreasing the dimensions

would reduce the computational effort but would also reduce the detail of the

image.

After the rotating step the image is converted to binary, filtered and then a

thinning method is applied in order to thin and repair the vein line. The ROI

treatment step can be seen in Figure 41.



Figure 41 - ROI treatment steps.

After being thinned, the standardized ROI is converted into a vector consisting in

luminance values that will be used in the feature extraction module.

Through the reference points illustrated on Figure 42, the value of 35 hand

geometry characteristics will be calculated in order to provide the geometrical

information of the hand. The characteristics used are the finger widths (20),

perimeters (5) and lengths (10). After acquiring the 35 hand geometry features, a

mean of the 35 values is calculated. This mean summarizes the geometrical

information of the hand, so each user in the database will have one mean

associated. At the identification stage, the mean of the recently acquired template

under identification will be compared with the remaining geometrical information

(means) of the previously acquired data in the database. Instead of comparing

templates randomly, the most probable will be compared first.

The most probable users will be the ones that have similar hand geometry. If the

vein pattern under identification does not fit the one from the user with the most

similar geometry, the algorithm searches the next most similar and so on, until

finding the one with the same vein pattern. The delay obtained by calculating the

hand geometry characteristics is almost irrelevant, due to the simple calculations

required.

The geometry similarity is not crucial for a positive matching, as it was already

explained, but helps sorting the most probable hands.



Figure 42 - Reference points used to calculate the hand geometry characteristics values.

4.2.3. Feature Extraction

The feature extraction module will output the biometric template, which will be

used in the matching stage. The feature extraction technique used in the developed

system is the Orthogonal Line Ordinal Features (OLOF) [5].

The vector of luminance values obtained in the preprocessing module will be the

input of the OLOF method that will generate a one bit feature code that is going to

be the template stored in the database.

The OLOF approach uses a 2D Gaussian filter to acquire the weighted average

intensity of a line-like region, equation (2)[5].

( ) ( (

( ) ( ) ( ) ( )

) (

( ) ( ) ( ) ( )

)

) (2)

In equation (2), symbolizes the orientation of the 2D Gaussian filter, the

filter’s horizontal scale and the filter’s vertical scale.

Equation (3) represents the orthogonal line ordinal filter, designed to compare two

orthogonal line-like palmprint image orientations for the same region [5]. The

image is turned into binary through the positive or negative output from the OLOF



equation. If the difference between the two Gaussian filters is positive corresponds

to 1, if it’s negative corresponds to 0.

( ) ( ) (

) (3)

The filtering of the ROI is accomplished using three orthogonal line ordinal filters

through three different orientations (θ), in this case: OF (0), OF (

) and OF (

). The

filter parameters used were and . The filter is centered at( )

( ). The output of the feature extraction phase using the OLOF extraction

method are three bit ordinal codes based on the sign of the filtering results (Figure

43 and Figure 44).

Figure 43- OLOF output in the three directions,

and with scale ratio equal to 3.

Figure 44 - OLOF output in the three directions, θ=π/6, θ=π/3 and θ=0 with scale ratio equal to 2.

The scale ratio (

) is controlled to be equal or higher than 3 to make the shape line

like [5]. The scale ratio used is 3 due to the results obtained in the Figure 45 and

Figure 46. From those figures it is clear that if a Scale Ratio equal to 2 is used, a

really low FAR is obtained but in other hand very high FRR values are acquired, so



it is not viable. The scale ratio equal to 4 in comparison with the scale ratio equal

to 3 obtains inferior FRR values but higher FAR results.

Figure 45 - FAR values for different scale ratios.

Figure 46 - FRR values for different scale ratios.

In order to reduce the computational complexity in the matching stage the three

resulting templates are resized. To find out which was the best template size, 300

different images with three different dimensions were used, 32x32, 64x64 and

128x128 for testing. Through the obtained results that are depicted in Figure 47, it

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

2

35 36 37 38 39

FAR

(%)

Threshold (%)

Scale Ratio 4

Scale Ratio 3

Scale Ratio 2

0

10

20

30

40

50

60

70

80

90

100

35 36 37 38 39

FRR

(%)

Threshold(%)

Scale Ratio 4

Scale Ratio 3

Scale Ratio 2



became obvious that the template size does not affect significantly the accuracy of

the system. The OLOF template size chosen is then 32x32, to reduce the

computational complexity.

Figure 47 - FAR versus FRR for different template sizes.

After the resizing stage, the images are saved in the database as a binary file

template in order to be used in the verification stage. A small extract from the

database is depicted in the Figure 48, where the 5 first lines represent 5 templates

from the same user and the lines 6 and 7 represent a different user.

Figure 48 - Database example.

0.00

5.00

10.00

15.00

20.00

25.00

0.00 0.05 0.10 0.15 0.20 0.25

FRR

(%)

FAR(%)

32x32

64x64

128x128



4.2.4. Matching System

A successful or unsuccessful recognition of an individual is based on the

calculation of the bitwise Hamming distances of the recently acquired template

and all the others in the database. The Hamming distance between two vectors is

the number of coefficients in which the corresponding symbols differ. If two

vectors are exactly equal, the Hamming distance will be zero. To calculate the

Hamming distance a bitwise XOR operator is used. The validation or refusal of the

matching is defined by a predefined threshold.



5. User Interface

In this section the palm vein recognition system implementation and the

developed graphical user interface are detailed.

This system is native to Matlab and is compiled to work on a Windows operating

system without needing Matlab installation, just a small pack of public libraries

that come with the executable file.

When a user starts the program, the initial interface menu will pop up, as

illustrated in Figure 49. The user is presented with four options that are explained

in detail in this section:

Identify: For users already registered in the system that want to be

identified. This is the default option to be selected.

Enroll: The user wants to get registered into the system’s database.

Send Database: The user wants to send the database to the database

manager.

Database Control: The user wants to test and correct the database to

eliminate possible errors.



Figure 49 - Initial Interface Menu

The first time the user runs the program, he must click in the “Enroll” button to

proceed with the registration, as shown in Figure 50.

Figure 50 - Enroll Interface



After filling the personal identification the user should start to acquire his

biometric characteristic by presenting the left hand to the system and pressing the

“Register Left Hand” button. To do this, the user must place his left hand inside the

box with his palm covering the entire illumination circle (Figure 51). The hand

position inside the box is crucial to a good image acquisition. The palm must cover

the circle because it is the thicker part of the hand and so it needs more

illumination intensity. During the capture process the user should move his hand

slightly in order to capture images in slightly different positions/angles. A total of

five images for each hand will be acquired.

Figure 51 - Correct Hand Placement

While doing the capture the hand contour is displayed in the user interface. That

contour can be red or yellow, meaning bad hand positioning and good hand

positioning respectively.

When the contour turns green it means that the image capture is completed. An

example of all the possible cases is presented in Figure 52.



Figure 52 –Three possible hand contours.

After finishing image acquisitions the algorithm will test them to conclude if they

were acquired correctly. If a problem occurs the system will prompt the user for a

new acquisition. Images may not be accepted due to different kinds of problems,

like noisy backgrounds and illumination disparities. An example of a corrupted

image is depicted in the Figure 53, where the bad hand placement led to a very

poor vein visibility.

Figure 53 - Example of a bad image acquisition.

To finish the enrollment of a hand, the system will ask the user to take the hand out

from the box and to insert it again in order to do a trial identification. If the user

name matches with the output displayed, the registration will be complete,

otherwise the system will ask for new images acquisition.



After finishing the acquisitions of the left hand the user must follow the same

protocol for the right hand.

After completing the acquisition for both hands, a green tick will appear in the

enroll interface menu and the user will be registered in the database.

After the registration process the user can push the “Identify” button to proceed

with the identification. If a match is found the username and his Facebook

photograph (obtained through the Facebook username) are displayed in the

matching result interface, Figure 54.

Figure 54 - Matching Result Interface

The “Send Database” button in the initial interface Menu sends the entire local

database to the database manager email in order to increment his database with

the registries of each person registered on the current computer.

The “Database Control” button was created in order to test the database for

possible errors. It tests if the database has the expected size for the number of

registered users. If an error is detected, the user has the option of repairing the



database, reenrolling all the users automatically. If the database is compromised,

the database control interface will display a red rectangle, Figure 55.

Figure 55 - Database Control Interface after detecting an error.

If there are no errors detected, a green rectangle is displayed, Figure 56.

Figure 56 - Database Control Interface after not detecting any error.



6. Experimental Results

On this chapter the results of the tests performed with the proposed biometric

systems are discussed. The system performance was tested using the commonly

used biometric performance measures mentioned in section 2.1.3 as references.

6.1. Database Creation

In order to test the performance of the developed system, the first step was to

create a hand palm vein database containing 30 registered people. For each

person, five different acquisitions from each hand were performed. For testing

purposes each hand is considered as a different user. 30 registered people

represent a database of 300 different templates.

6.2. Performance Evaluation

The performance of the developed biometric system is evaluated by the ROC curve

which plots the FAR against the GAR (or 1-FRR). The performance is also evaluated

by the Equal Error Rate (EER), which is defined as the error rate when the FAR and

the FRR are equal. A recognition attempt might have the following results:

Type of user Match Non-Match

Genuine Correct Accept False Reject

Impostor False Accept Correct Reject



The system performance was tested using the database mentioned above, using a

ROI size of 128x128 and the features were extracted with the OLOF technique. The

FAR, FRR and GAR results obtained with the developed system are presented at

Table 2.

Threshold (%)

FAR (%) FRR (%) GAR (%)

0 0,00 100,00 0,00

5 0,00 100,00 0,00

10 0,00 99,83 0,17

15 0,00 96,50 3,50

20 0,00 83,50 16,50

25 0,00 56,00 44,00

30 0,00 33,00 67,00

35 0,00 19,67 80,33

40 0,51 13,33 86,67

45 20,97 6,83 93,17

50 88,13 0,00 100,00

55 99,98 0,00 100,00

60 100,00 0,00 100,00

65 100,00 0,00 100,00

70 100,00 0,00 100,00

75 100,00 0,00 100,00

80 100,00 0,00 100,00

85 100,00 0,00 100,00

90 100,00 0,00 100,00

95 100,00 0,00 100,00

100 100,00 0,00 100,00

Table 2 - FAR, FRR and GAR for different threshold values.

The obtained ROC curve is shown in Figure 57. The ROC curve is near the perfect

point (0,100) which shows the good matching performance of the system.

The ROC curve and the table show that GAR is near 85% when the FAR is 0%.



For applications like opening doors for not very high secure areas, an operation

point with values of FAR above 0% can be used despite the slope of the ROC curve

suggesting that the GAR increases very slowly in comparison with the FAR, so

there is no great benefit on using a GAR above 85%.

For ATM machine operations or Internet bank account managements the FAR must

be near 0%, which will result in an operating point leading to GAR values below or

equal to 85%.

Figure 57 - Receiver Operation Characteristic curve for a ROI with 128x128 pixels.

An alternative and simplified way of evaluating the performance of a biometric

system is through the EER. A low EER means that it is possible to get both low

values of FRR and of FAR and thus the lower the EER, the better the performance.

Despite being a good reference point, the EER might not be the ideal operating

point for a given system. The system might require a lower FRR or FAR for special

application conditions. A system that requires high security conditions, such as the

ATM machine, will require a really low FAR which will possibly imply a higher

FRR.

0

10

20

30

40

50

60

70

80

90

100

0 20 40 60 80 100

GA

R(%

)

FAR(%)

ROC

ROC



Figure 58 shows the FAR and FRR curves produced in this dissertation as functions

of the threshold. The figure shows that when the threshold value increases, the

FRR decreases and the FAR increases. The figure also shows that if the threshold is

lower that 40% the FAR is near zero. Through the figure it is perceptive that the

EER of the developed system is near 9% and the associated threshold is about

45%.

Figure 58 - FAR and FRR at different operating thresholds.

In order to test which ROI size should be used, three ROC curves were created. The

three sizes tested were, 32x32, 64x64 and 128x128 pixels. The obtained result is

depicted in Figure 59.

0

10

20

30

40

50

60

70

80

90

100

0 10 20 30 40 50 60 70 80 90 100

%

Threshold (%)

FAR(%)

FRR(%)



Figure 59 - ROC curve for different ROI dimensions.

From the Figure 59, it is obvious that the ROI size of 128x128 pixels and 64x64

obtain the best results in terms of matching. The ROI size of 32x32 pixels clearly

underperforms both in the ROC curve as well in the EER (see Figure 60).

Figure 60 - FRR (%) against FAR (%) to obtain EER for different ROI dimensions.

0

10

20

30

40

50

60

70

80

90

100

0 0.1 0.2 0.3 0.4 0.5

GA

R(%

)

FAR(%)

128x128

64x64

32x32

0

5

10

15

20

25

30

35

40

45

50

0 10 20 30 40 50

FRR

(%)

FAR(%)

128x128

64x64

32x32

EER



The ROI size chosen for the developed system is 128x128 pixels due to the better

matching performance.

6.3. Operating Point Selection

The operating point used depends on the application. It must be chosen taking into

account the system recognition performance and the security of the system. The

developed system is intended to be used in several applications, as it is mentioned

in section 4.1. The operating points present on Table 3 were obtained with a

database of 30 different users, and might have slightly changes if the number of

users is increased.

Threshold (%)

FAR (%) FRR (%)

35 0,000 19,667 35,5 0,002 18,833 36 0,009 18,167

36,5 0,011 18,000 37 0,020 17,500

37,5 0,038 16,167 38 0,072 15,167

38,5 0,113 14,667 39 0,199 14,000

39,5 0,337 13,833 40 0,508 13,333

Table 3 - Values of FAR and FRR for different operating points.

If the system is intended to be used on an ATM machine or to do online bank

account operations, the operating point should be the one depicted in red. That

operating point implies that in every 1000 attempts to access the system 0

impostors (0.0%) will be accepted and around 197 genuine users (19,66%) will be

rejected which means that they require a new authentication attempt.



If the purpose of the developed system is opening a garage door, the operating

point chosen should be the one depicted in green on the Table 3. Through this

operation point in every 1000 attempts to access the system, 5 impostors (0.5%)

will be accepted and around 133 genuine users (13.33%) will be rejected,

requiring a new authentication attempt.

If the application was intended to unlock a computer or access to some secure

folders, the operating point depicted in blue in the Table 3 could be used. This

operating point provides that in 1000 attempts 2 impostors (0.2%) will be

accepted and 140 genuine users (14%) will be rejected, requiring a new

recognition attempt.

It is important to highlight that the databases behind the results of each technique

discussed below are not the same, which limits their comparability. However,

these results are used as guiding points to evaluate the developed technique.

Mauricio Ramalho [16] in his palmprint recognition system used an operating

point that achieves 9.5% for the FRR, 0.1% for FAR and 3.29% for the EER. The

system proposed in this dissertation achieves slightly worst results, but has all the

advantages associated with the vein pattern over the palm print pattern

(invisibility expect under special conditions, being internal to the human body, and

providing liveliness prove). Also, Mauricio Ramalho [16] used an Olympus C-3020

Z digital camera that is clearly more expensive than the web-camera used in the

proposed thesis (Logitech QuickCam Pro 9000).

Nuno Moço [23] in his palmprint recognition system for cellphones used an

operating point that achieved 9.87% of FRR and 0.03% of FAR with an EER of

around 5%.

Huan Zhang and Dewen [2] on theirs hand vein recognition system achieved an

EER of 1.82% with an AD-080CL camera that costs around 3000 €, which is 120

times more expensive than the web-camera of the proposed system (25 €).



7. Plans for the Future

During the progress of this thesis an alternative system was also created. It had the

purpose of using the hand texture pattern as the identification factor, in non-

controlled environments. Due to time constraints it was not possible to turn this

system robust enough.

The hand texture pattern recognition system was intended to be used to:

Unlock Computer – Instead of choosing the user name and password to do the

computer login a brief hand palm scan should be sufficient to associate the data

obtain to a specific user.

Unlock Cellphone – In a similar way as the approach taken before, a cell phone

version might be developed in order to do a cell phone unlock.

This system uses a similar algorithm as the vein pattern recognition system with

some improvements. The main difference is the usage of a background subtraction

algorithm, which is used to simplify the hand detection phase. The background

subtraction algorithm turns all the pixels inside the areas without movement black.

The hand detection phase is accomplished with the help of a skin detection

algorithm that searches the images for skin color pixels. The hand detection phase

can be seen in Figure 61.

Figure 61 - Hand texture recognition system using a regular laptop computer camera.



The main objective in the near future is turning the hand feature recognition

system more robust and combining it with the hand vein recognition system on a

special assembly.

The developed hand veins recognition system also requires some development in

order to become even more robust:

The developed box should be painted in black in order to reduce the

amount of visible light present inside the box during the acquisitions.

It should be possible to automatically regulate the IR illumination intensity

inside the box in order to obtain good results for hands with all sizes and

thicknesses.

The system should be bimodal and require a hand geometry matching for a

valid authentication.

The database size should be larger in order to test the system performance

more accurately.



8. Conclusions

This dissertation presents a unimodal biometric recognition system that used the

hand vein patterns to do the identification of an individual. It was developed in

Matlab and implemented to work on a Windows operation system.

The developed system has proved to work well for the application scenarios

considered, such as ATM operations, opening doors and unlocking computers. In

addition, it has the advantage of being low-cost, requiring an investment of around

50€ and of being simple to assemble in comparison with the existing recognition

systems, as discussed throughout this dissertation.

Through testing, several operating points were obtained and associated to

different applications, with the objective of obtaining the best response from the

system in terms of recognition and security performances for the different

purposes of use. The EER of the developed system is near 9%. The ROI dimension

used is 128x128 pixels due to the best matching results during the tests. The OLOF

templates dimensions used in order to provide a smaller database without losing

performance is 32x32.

The OLOF approach used has proved to be very effective at providing good

biometric recognitions results for the tested database and in the live acquisitions

as discussed in chapter 4.2.3.



9. References

[1] Greitans, M., Pudzs, M. & Fuksis, R., 2010, “Palm Vein Biometrics Based on

Infrared Imaging and Complex Matched Filtering”, Proceedings of the 12th

ACM Workshop on Multimedia and Security, pp.101-106, New York, USA.

[2] Zhang, H. & Hu, D., 2010, “A Palm Vein Recognition System”, IEEE

Proceedings of the International Conference on Intelligent Computation

Technology and Automation, pp.285-288.

[3] The International Biometric Group – available at:

http://www.biometricgroup.com/, accessed on 2013/11/11.

[4] Edgington, B., 2007 “Introducing Hitachi’s Finger Vein Technology-A

White Paper”, Hitachi’s Finger Vein Technology, Version 1.0.

[5] Sun, Z., Tan, T., Wang, Y., 2005, “Ordinal Palmprint Representation for

Personal Identification”, IEEE Proceedings of the Computer Vision and

Pattern Recognition, 1, pp.279-284.

[6] Canny, J., 1986, “A computational approach to edge detection”, IEEE

Transactions on Pattern Analysis and Machine Intelligence, pp.450–463.

[7] Wu, X., Wang, K., Zhang, D., 2002, “Line feature extraction and matching in

palmprint”, Proceeding of the Second International Conference on Image

and Graphics, pp.583–590.

[8] Palm Print ATM at Poland - available at:

http://blog.antivirus365.net/?p=323, accessed on 2012/12/13.

[9] AOptix InSight Duo used at Airports - available at:

http://arstechnica.com/business/2012/09/company-bets-on-airport-of-

the-future-passing-security-with-an-iris-scan/, accessed on 2013/02/04.

[10] AOptix InSight Duo technology – available at:

http://www.aoptix.com/identity-solutions/high-

throughput/products/insight-duo/, accessed on 2013/05/20.

http://www.biometricgroup.com/

http://blog.antivirus365.net/?p=323

http://arstechnica.com/business/2012/09/company-bets-on-airport-of-the-future-passing-security-with-an-iris-scan/

http://arstechnica.com/business/2012/09/company-bets-on-airport-of-the-future-passing-security-with-an-iris-scan/

http://www.aoptix.com/identity-solutions/high-throughput/products/insight-duo

http://www.aoptix.com/identity-solutions/high-throughput/products/insight-duo



[11] Passwords Weaknesses – available at:

http://www.pcworld.com/article/207718/surprise_passwords_are_still_

weak_link_in_security_chain.html accessed on 2013/11/15.

[12] Jain, A., Ross, A. & Prabhakar, S., 2004 “An Introduction to Biometric

Recognition”, IEEE Transactions on Circuits and Systems for Video

Technology, 14, pp.4-20.

[13] Shapiro, L. & Stockman, G., 2000, “Computer Vision”, Prentice Hall, Upper

Saddle River, N.J.

[14] Konukoglum, E., Yorukm, E., Darbon, J. & Sankurm, B., 2006, “Shape-Based

Hand Recognition”, 2006, IEEE Transactions on Image Processing, 15 (7),

pp.1803-1815.

[15] Lin, C., Chuang, T. & Fan, K., 2005, “Palmprint Verification using

Hierarchical Decomposition”, Pattern Recognition, 38 (12), pp.2639-2652.

[16] Ramalho, M., 2010, “Secure Palmprint Verification System”, Master Degree

Dissertation, Instituto Superior Técnico, Lisboa.

[17] Vision-box developed system – available at: http://www.vision-

box.com/solutions/vb-i-match/ accessed at 2013/05/02.

[18] Connie, T., Jin, A., Ong, M. & Ling, D., 2005, “An Automated Palmprint

Recognition System” , Image and Vision Computing, 15 (5), pp.501-515.

[19] Kumar, A., Wong, D.C.M., Shen, H.C. & Jain, A.K., 2003, “Personal

Verification using Palmprint and Hand Geometry Biometric”, in

Proceedings of the 4th International Conference on Audio and Video Based

Biometric Person Authentication, Guildfrd, U.K, pp.668-678.

[20] High Power Infrared Emitter datasheet - available at:

http://www.farnell.com/datasheets/1441891.pdf accessed on

2013/03/12.

[21] Soille, P., 1999, “Morphological Image Analysis: Principles and

Applications”, Springer-Verlag, pp 173-1748.

[22] Niblack, W., 1986, “An introduction to digital image processing”, Prentice

Hall Englewood Cliffs, NJ, pp.115-116.

http://www.vision-box.com/solutions/vb-i-match/

http://www.vision-box.com/solutions/vb-i-match/

http://www.farnell.com/datasheets/1441891.pdf



[23] Moço, N., 2012, “Biometric Recognition Based on Smartphone”, 2012,

Master Degree Dissertation, Instituto Superior Técnico, Lisboa.

[24] Duta, N., Jain, A. & Mardia, K., 2002, “Matching of Palmprint”, Pattern

Recognition Letters, 23, pp.477-485.

[25] Huang, D., Jia, W. & Zhang, D., 2008, “Palmprint verification based on

principal lines”, Pattern Recognition”, 41 (4), pp.1316–1328.

[26] You, J., Li, W. & Zhang, D., 2002, “Hierarchical palmprint identification via

multiple feature extraction”, Pattern Recognition, 35 (4), pp.847–859.

[27] Michael, G., Connie, T., & Teoh, A., 2010, “A Contactless Biometric System

Using Palm Print and Palm Veins Features”, online publication.

[28] Wu, X., Zhang, D. & Wang, K., 2003, “Fisher palms based palmprint

recognition”, Pattern Recognition Letters, 24, (15), pp.2829–2838.

[29] Funada, J., Ohta, N., Mizoguchi, M., Temma, T., Nakanishi, T., Murai, K., et al.,

1998, “Feature extraction method for palmprint considering elimination of

creases”, Proceedings of the 14th International Conference of Pattern

Recognition, 2, pp.1849-1854.

[30] Wu, X., Wang, K. & Zhang, D., 2004, “A novel approach of palm-line

extraction”, Proceedings of the Third International Conference on Image and

Graphics, pp.230–233.

[31] Diaz, R., Travieso, M., Alonso, C. & Ferrer, J., 2004, “Biometric system based

in the feature of hand palm”, Proceedings of 38th Annual International

Carnahan Conference on Security Technology, pp.136–139.

[32] Kung, S., Lin, S. & Fang, M., 1994, “A neural network approach to face/palm

recognition”, Proceedings of IEEE Workshop on Neural Networks for Signal

Processing, pp.323–332, Cambridge.

[33] Huang, D., Jia, W. & Zhang, D., 2008, “Palmprint verification based on

principal lines”, Pattern Recognition, 41, (4), pp.1316–1328.

[34] Lu, G., Zhang, D. & Wang, K., 2004, “Palmprint recognition using Eigen

palms features”, Pattern Recognition Letters, 24 (9), pp.1463–1467.



[35] Connie, T., Jin, A., Ong, M. & Ling, D., 2005, “An automated palmprint

recognition system”, Proceedings of Image and Vision Computing, 23 (5),

pp. 501–515.

[36] Sun, Z., Tan, T., Wang, Y. & Li, S., 2005, “Ordinal palmprint representation

for personal identification”, Proceedings of Computer Vision and Pattern

Recognition, 1, pp.279–284.

[37] “The History of Fingerprints” – available at:

http://onin.com/fp/fphistory.html accessed on 2013/05/14.

[38] “The Impact of Biometrics” - available at:

http://www.le.ac.uk/oerresources/criminology/msc/unit8/page_19.htm

accessed at 2012/12/15.

[39] International Biometrics & Identification Association, 2013, “Biometrics

and Identity in the Digital World”, pp.1-3.

[40] “Hitachi biometric systems around the world” - available at:

http://www.hitachi.com/New/cnews/120206b.html accessed at

2013/06/03.

[41] Bansal, A., Agarwal, R. & Sharma, R.K., 2010, “Trends in Iris Recognition

Algorithms”, Proceedings of the Fourth Asia International Conference on

Mathematical/Analyrical Modeling and Computer Simulation, pp.337-340.

[42] Ong, M., Tee, C. & Jin, A., 2008, “Touch less Palm Print Biometric”,

Proceedings of the International Conference on Computer Vision Theory and

Applications, 26 (12), pp.1551-1560.

[43] Otsu, N., 1979, “A Threshold Selection Method from Gray-Level

Histograms”, IEEE Transactions on Systems, Man, and Cybernetics, 9 (1),

pp.62-66.

http://onin.com/fp/fphistory.html

http://www.le.ac.uk/oerresources/criminology/msc/unit8/page_19.htm

http://www.hitachi.com/New/cnews/120206b.html%20accessed%20at%202013/06/03

http://www.hitachi.com/New/cnews/120206b.html%20accessed%20at%202013/06/03

hand veins recognition system - fenix.tecnico.ulisboa.pt · hand-based biometrics, biometrics...

Documents