sharma thesis1

M.E.DIGITAL COMMUNICATION

PIET Page i

ACKNOWLEDGEMENT

I, hereby, take an opportunity to convey my gratitude for the generous

assistance and cooperation, that I received from the PG Director Dr. P.H Tandel and

to all those who helped me directly and indirectly.

I am sincerely thankful to my Guide, Asst .Prof. Mitul M. Patel whose

constant help, stimulating suggestions and encouragement helped me in completing

my Literature Review work successfully.

I am also thankful to Prof. M.A.Lokhandwala, Head of the Department and

other faculty members who have directly or indirectly helped me whenever it was

required by me.

I am thankful to Prof. Jitendra Chaudhary and Prof. Sankar Parmar who

helped me and encouraged me in my work.

Finally, I am also indebted to God and my friends without whose help I would

have had a hard time managing everything on my own.

Sharma Ashok Sukhbir

[110370722003]


PIET Page ii

TABLE OF CONTENTS

ACKNOWLEDGEMENT ............................................................................................. i

TABLE OF CONTENTS ............................................................................................. ii

LIST OF FIGURE ....................................................................................................... iv

LIST OF TABLES ....................................................................................................... vi

ABSTRACT vii

CHAPTER 1 INTRODUCTION .................................................................................... 1

1.1 OVERVIEW OF BIOMETRICS ..................................................................... 2

1.2 PALM PRINT BIOMETRIC RECOGNITION............................................... 4

1.3 HISTORY ........................................................................................................ 5

1.4 DESCRIPTION AND OUTLINE OF THE THESIS ...................................... 7

CHAPTER 2 BIOMETRICS IN AUTHENTICATION ................................................. 9

2.1 INTRODUCTION ......................................................................................... 10

2.2 PROPERTIES OF BIOMETRICS ................................................................. 10

2.3 BIOMETRIC SYSTEM BLOCK DIAGRAM .............................................. 11

2.4 VERIFICATION AND IDENTIFICATION ................................................. 12

2.5 LEADING BIOMETRIC TECHNOLOGIES ............................................... 13

2.5.1 FINGER-SCAN .................................................................................. 13

2.5.2 FACIAL-SCAN .................................................................................. 14

2.5.3 IRIS-SCAN ......................................................................................... 14

2.5.4 VOICE-SCAN .................................................................................... 15

2.6 DESIRED FEATURES IN A BIOMETRIC ................................................. 16

CHAPTER 3 EXISTING PALMPRINT RECOGNITION ALGORITHMS ................ 18

CHAPTER 4 PROPOSED SYSTEM ............................................................................ 22


PIET Page iii

4.1 ROI EXTRACTION ...................................................................................... 24

4.1.1 BINARIZATION .............................................................................. 24

4.1.2 CONTOURING ................................................................................ 25

4.1.3 SELECTING REFERANCE POINT ................................................ 26

4.1.4 CROPING ROI ................................................................................. 27

4.2 ENHANCEMENT OF ROI ........................................................................... 27

4.3 FEATURE EXTRACTION AND CODING ................................................. 29

4.4 ZERNIKE MOMENTS ................................................................................. 31

CHAPTER 5 RESULTS ............................................................................................... 35

CHAPTER 6 CONCLUSION AND FUTURE SCOPE ................................................ 44

6.1 CONCLUSION .............................................................................................. 45

6.2 FUTURE SCOPE........................................................................................... 45

REFERENCES ............................................................................................................. 46

APPENDIX ……………………………………………………………………….....50

APPENDIX-A .............................................................................................................. 51


PIET Page iv

LIST OF FIGURE

Figure 1.1:Biometric characteristics .............................................................................. 3

Figure 1.2:Typical scheme of a biometric system. ........................................................ 4

Figure 2.1:Block Diagram of the Proposed Algorithm ................................................ 11

Figure 3.1:Schematic Diagram of Palmprint Acquisition System [7] ......................... 20

Figure 3.2:Pegs and the Cropped Area of the Palm [7] ............................................... 20

Figure 4.1:Block diagram of palm print verification system ....................................... 23

Figure 4.2:Database examples ..................................................................................... 24

Figure 4.3:Binarized image .......................................................................................... 25

Figure 4.4:Palm print contour ...................................................................................... 26

Figure 4.5:Distance transform of contour image ......................................................... 26

Figure 4.6:Region to be cropped .................................................................................. 27

Figure 4.7:ROI ............................................................................................................. 27

Figure 4.8:(a) palm print ROI (b) coarse reflection, (c) uniform brightness palm print

image, (d) Enhanced palm print image ........................................................................ 28

Figure 4.9:Principle Lines and Wrinkles in a Palm [20] ............................................. 30

Figure 4.10:Three Sets of Palmprint Images with Similar Principal Lines from

Different People ........................................................................................................... 31

Figure 4.11:Square to circular transform. .................................................................... 33

Figure 5.1:Minimum distance for test image 8 ............................................................ 36





Figure 5.6:Minimum distance for test image 11 .......................................................... 38

Figure 5.7:Value of minimum distance for test image 20 ........................................... 39

Figure 5.8:Train index values for corresponding test images ...................................... 39

Figure 5.9:Minimum distance graph for all test images .............................................. 40

Figure 5.10:False matched images ............................................................................... 40

Figure 5.12:Result of thresholding .............................................................................. 41

Figure 5.13:Smallest distance histogram ..................................................................... 42


PIET Page v

Figure 5.14:Second smallest distance histogram ......................................................... 42

Figure 5.15:Reliability of identification ...................................................................... 43


PIET Page vi

LIST OF TABLES

Table 4.1: DPI REQUIREMENTS ....................................................................................... 30

Table 5.1: EFFICIENCY....................................................................................................... 43


PIET Page vii

ABSTRACT

Biometrics is identification of an individual on the basis of unique physiological and

behavioral patterns. Biometrics is fast replacing other means of authentication like

passwords and keys due to the inherent drawbacks in them and increased

effectiveness and reliability of the biometric modalities. The passwords can be

forgotten or hacked, while keys can be lost. The individual’s unique physiological or

behavioral characteristics, on the other hand, are hard to forged or lost. Finger print

and face are the common biometrics used nowadays, but they have inherent problems.

The illumination variations affect the performance of face recognition algorithms,

while finger print, along with technological challenges, has less user acceptability due

to the historical use in crime investigations. Palm print is a biometric modality which

has recently drawn great attention owing to its strengths like ease of acquisition,

robustness, user acceptance in addition to its uniqueness and rich distinguishable

contents and features. Palm print biometric is potentially a very effective biometrics

in sense it offers widely discernible and discriminating features like palm lines,

wrinkles, minutiae and delta points.

Limited work has been reported on palm print based identification/verification despite

of its significant features. Efforts have been made to build a palm print based

recognition system based on structural features of palm print like crease points, line

features, Datum points, local binary pattern histograms. There also exists systems

based on statistical features of palm print extracted using Fourier transforms, Discrete

Cosine Transforms, Karhunen-Lowe transforms, Wavelet transforms, Independent

Component Analysis, Gabor filter, Linear Discriminant Analysis(LDA), Neural

networks, statistical signature and hand geometry.

In current work, the palm print identification is done with the help of a statistical

method, Moments. Moments are used from early times in image processing for

character recognition and statistical analysis. Using this technique in biometrics

induces some benefits like invariance in rotation and translation. The moments are

used as features and its extraction and matching will be done using the software

MATLAB.


PIET Page 1

CHAPTER 1

INTRODUCTION


PIET Page 2

1.1 OVERVIEW OF BIOMETRICS

Studies of physical and behavioral traits are known as biometrics. Being specific of an

individual, they guarantee one’s identity in security control situations. A well-known

example of a biometric characteristic are fingerprints, which is the most widely used.

It is theoretically impossible to find any two individuals with the same fingerprint.

This is a crucial property of a biometric characteristic: to be unique for each person.

Other equally important aspects regarding biometric characteristics are universality,

as they have to be present in all individuals; and permanence, so they are constant

during one’s life. Moreover, they should be easy to extract. At present time, there is

research activity in a broad range of biometric characteristics which can be divided

into physical and behavioral. Physical are, for instance, fingerprints, iris, retinal

capillary structure, face, and hand recognition. Examples of behavioral traits are voice

and handwriting. Figure 1illustrates several biometric characteristics.

Biometric systems can be used for identification and recognition purposes. In all cases

there should be a database where biometric features from a set of individuals are

stored. The role of the system is to compare an input with all the entries in the

database and verify if there is a match, thus confirming the identity of the individual.

To compare any kind of biometric characteristics it is necessary to represent them in a

stable fashion. For instance, it is not feasible to directly compare images from two

palm prints, as it is practically impossible to place the hand in the exact same position

in different occasions, producing slightly different images that have to be compared

income way.

This is the most crucial aspect and can be divided into two tasks:

1. Represent a biometric characteristic in reproducible and stable features

that resist input variability.

2. Compare such features so users can accurately be identified.

These two questions are in the core of a biometric system and are addressed by most

of the researching the field. Its importance is highlighted in figure 2 where the layout

of a biometric system is depicted.


PIET Page 3

Figure Error! No text of specified style in document..1:Biometric characteristics

a) Retinal fundus image and b) detection of correspondent vascularization. c) and d) are examples of

retinal processed images from different persons. e) and f) are fingerprints from twin sisters, with

noticeable differences to the naked eye. g) represents the pressure-time plot of the utterance(complete

unit of speech), from which spectral information can be retrieved to identify a speaker. h) and i) depict

the illumination system for acquisition of 3D palm print information, resulting in j). k) and l) are iris

from two different persons. m) depicts palm veins detections using infra-red lighting.


PIET Page 4

Figure 1.2:Typical scheme of a biometric system.

1.2 PALM PRINT BIOMETRIC RECOGNITION

During the last years there has been an increasing use of automatic personal

recognition systems. Palm print based biometric approaches have been intensively

developed over the last 12 years because they possess several advantages over other

systems.

Palm print images can be acquired with low resolution cameras and scanners and still

have enough information to achieve good recognition rates. If high resolution images

are captured, ridges and wrinkles can be detected. Forensic applications typically

require high resolution imaging, with at least 500 dpi.

According to the classification in, palm prints are one of the four biometric modalities

possessing all of the following properties:

• Universality, which means, the characteristic should be present in all

individuals;

• Uniqueness, as the characteristic has to be unique to each individual;

• Permanence: its resistance to aging;

• Measurability: how easy is to acquire image or signal from the individual;

Acquisition Feature

Extraction

Database

Feature

Matching

User input User registration

System

output

Verification

Identification


PIET Page 5

• Performance: how good it is at recognizing and identifying individuals;

• Acceptability: the population must be willing to provide the characteristic;

• Circumvention: how easily can it be forged;

The other three modalities are fingerprints, hand vein and ear canal. For instance, iris

based methods, which are the most reliable; require more expensive acquisition

systems than palm print systems. Face and voice characteristics are easier to acquire

than palm prints, but they are not so reliable. Overall, palm print based systems are

well balanced in terms of cost and performance.

1.3 HISTORY

In many instances throughout history, examination of handprints was the only method

of distinguishing one illiterate person from another since they could not write their

own names. Accordingly, the hand impressions of those who could not record a name

but could press an inked hand onto the back of a contract become an acceptable form

of identification. In 1858, Sir William Herschel, working for the civil service of India,

recorded a handprint of back of a contract for each worker to distinguish employees

from others who might claim to be employees when payday arrived. This was the first

recorded systematic capture of hand and finger images that were uniformly taken for

identification purposes.

The first known AFIS system built to support palm print is believed to have been built

by a Hungarian company. In late 1994, latent experts from the United States

benchmarked the palm system and invited the Hungarian company to the 1995

International Association for Identification (IAI) conference. The palm system was

subsequently bought by a US company in 1997.

In 2004, Connecticut, Rhode Island and California established state wide palm print

databases that allowed law enforcement agencies in each state to submit unidentified

latent palm prints to be searched against each other’s database of known offenders.

Australia currently houses the largest repository of palm prints in the world. The new

Australian Nation Automated Fingerprint Identification system (NAFIS) includes 4.8


PIET Page 6

million palm prints. The new NAFIS complies with the ANSI/NIST international

standard for fingerprint data exchange, making it easy for Australian police services

to provide finger print records to overseas police forces such as Interpol or the FBI,

when necessary.

Over the past several years, most commercial companies that provide fingerprint

capabilities have added the capability for storing and searching palm print records.

While several state and local agencies within the US have implemented palm systems,

a centralized nation palm system has yet to be developed. Currently, the Federal

Bureau of Investigation (FBI) Criminal Justice Information Services (CJIS) Division

houses the largest collection of criminal history information in the world. This

information primarily utilizes fingerprints as the biometric allowing identification

services to federal, state, and local users through the Integrated Automated

Fingerprint Identification system (IAFIS). The Federal Government has allowed

maturation time for the standards relating to palm data and live-scan capture

equipment prior to adding this capability to the current services offered by the CJIS

Division. The FBI Laboratory Division has evaluated several different commercial

palm AFIS systems to gain a better understanding of the capabilities of various

vendors. Additionally, state and local law enforcement have deployed systems to

compare latent palm prints against their own palm print databases. It is a goal to

leverage those experiences and apply them towards the development of a National

palm Print Search System.

In April 2002, a Staff Paper on palm print technology and IAFIS palm print

capabilities were submitted to the identification Services (IS) Subcommittee, CJIS

Advisory Policy Board (APB). The Joint Working Group then moved “for strong

endorsement of the planning, costing, and development of an integrated latent print

capability for palms at CJIS Division of the FBI. This should proceed as an effort

along the same parallel lines that IAFIS was developed and integrate this into the

CJIS technical capabilities...”

As a result of this endorsement and other changing business needs for law

enforcement, the FBI announced the Next Generation IAFIS (NGI) initiative. A major

component of the NGI initiative is the development of the requirements for and


PIET Page 7

deployment of an integrated NATIONAL Pam print service. Law enforcement

agencies indicate that at least 30 percent of the prints lifted from crime scenes – from

knife hilts, gun grips, steering wheels, and window panes – are of palms, not fingers.

For this reason, capturing and scanning latent palm prints is becoming an area of

increasing interest among the law enforcement community. The improving law

enforcement’s ability to exchange a more complete set of biometric information,

making additional identifications, quickly aiding in solving crimes that formerly may

have not been possible, and improving the overall accuracy of identification through

the IAFIS criminal history records.

1.4 DESCRIPTION AND OUTLINE OF THE THESIS

In this work, a palmprint recognition algorithm based on Zernike moments is

implemented in MATLAB environment. Nevertheless, built-in functions in

MATLAB® Image Processing Toolbox are almost not utilized in order to develop a

platform-independent algorithm. The developed algorithm is first tested on The Hong

Kong Polytechnic University Palmprint Database.

This thesis is organized as follows: In Chapter II, biometrics, the emerging and

reliable authentication method, is discussed in detail. In this chapter, key metrics used

in the evaluation of biometric systems have been defined, and advantages and

disadvantages of some leading biometric technologies currently being used are

mentioned. Moreover, advantages of palmprint as a biometric have been explained.

In Chapter III, brief information is given about The Hong Kong Polytechnic

University Palmprint Database, the most commonly used palmprint database which is

also used in this thesis. Furthermore, some of the palmprint recognition methods in

the literature are described and results obtained in these studies are presented.

In Chapter IV details of the developed palmprint recognition algorithm are given. In

this chapter, the algorithm is divided into three sub-blocks and each sub-block is

detailed along with the discussions related to the effects of different parameters.


PIET Page 8

In Chapter V, results of the developed algorithm on The Hong Kong Polytechnic

University Palmprint Database are presented. Obtained results are investigated from

different side of views and factors affecting results are discussed. Moreover, some

strong and weak points of the algorithm together with some possible improvements on

palmprint recognition system are discussed in Chapter VI.


PIET Page 9

CHAPTER 2

BIOMETRICS IN AUTHENTICATION


PIET Page 10

2.1 INTRODUCTION

Because biometric based authentication is emerging as a powerful method for reliable

authentication, which is of great importance in our lives, biometrics is becoming

increasingly popular. In 2001, the highly respected MIT Technology Review

announced biometrics as one of the “top ten emerging technologies that will change

the world” [1]. Also Rick Norton, the executive director of the International Biometric

Industry Association (IBIA), pointed out the increase in biometric revenues by an

order of magnitude over the recent years. Biometric revenues, which were $20 million

in 1996, increased by 10 times and reached $200 million in 2001. Rick Norton

expects a similar increase in biometric revenues in next 5 years period, from 2001 to

2006, thereby expecting them to reach $2 billion by 2006[1]. Similarly, International

Biometric Group, a biometric consulting and integration company in New York City,

estimate biometric revenues to be around $1.9 billion in 2005[1].

2.2 PROPERTIES OF BIOMETRICS

Researchers noticing the increase in biometric revenues are trying to develop better

algorithms for existing biometrics and\or to find new biometrics for authentication.

Whether new or existing, all practical biometrics should possess five properties

described below [2]:

1. Universality: All individuals should possess the biometric characteristics.

2. Uniqueness: The biometric characteristics of different individuals should

not be the same.

3. Permanence: The biometric characteristics of individuals should not change

severely with the time.

4. Collectability: The biometric characteristics should be measurable with

some practical device.

5. Acceptability: Individuals should not have objections to the measuring or

collection of the biometric.


PIET Page 11

2.3 BIOMETRIC SYSTEM BLOCK DIAGRAM

Figure 2.1:Block Diagram of the Proposed Algorithm

After the biometric that is to be utilized is decided, the question how a biometric

system can be implemented naturally arises. Figure 2.1 shows the general block

diagram of a biometric system. As shown in Figure 2.1, biometric systems generally

consist of the following components:

Data Acquisition Block: This is the block in which biometric data is

captured and is transferred to feature extraction and coding block. The

biometric data may also be compressed in this block, especially when

the data acquisition is performed at a remote location.

Transmission Channel Block: This is an optional block in the sense

that some biometric systems do not consist of this block. Although

transmission channels are internal to the device in self-contained


PIET Page 12

systems, some biometric systems may be distributed and may have

central data storage and many remote data acquisition points. The

transmission channel for distributed systems might be a local area

network (LAN), a private Intranet, or even the Internet. [1]

Feature Extraction and Coding Block: This is the block in which

acquired biometric sample is processed. Processing consists of

segmentation, the process of separating relevant biometric data from

background information, and feature extraction, the process of locating

and extracting desired biometric data. After segmentation and feature

extraction, a biometric template, a mathematical representation of the

original biometric, is obtained by encoding extracted features.

Distance Matching and Decision Policy Block: This is the final block

in a biometric system, where the final decision is made. The biometric

template obtained in feature extraction and coding block is compared

to one or more templates in the data storage by selected matching

algorithm, which determines the degree of similarity between

compared templates. The final decision is usually made based on the

result of the matching algorithm and empirically determined

thresholds.

2.4 VERIFICATION AND IDENTIFICATION

The most important distinction in biometrics is between verification and

identification. Verification systems verify or reject users’ identity. In verification

systems, the user is requested to prove that he/she is the person he/she claims to be.

Therefore; the user should first claim an identity by providing a username or an ID

number. After claiming the identity, the user provides a biometric data to be

compared against his or her enrolled biometric data. The biometric system then

returns one of two possible answers, verified or not verified. Verification is usually

referred to as 1:1 (one-to-one), since the biometric data provided by the user is only

compared against the enrolled biometric data of the person that the user claims to be.

Identification systems, on the other hand, try to identify the person providing the

biometric data. In identification systems, the user is not required to claim an identity;

which is not the case in verification systems, instead he/she is only requested to


PIET Page 13

provide a biometric data. Another difference of identification from verification is that

user’s biometric data is compared against a number of users’ biometric data.

Therefore; identification is generally referred as 1:N (one-to-N or one-to-many). Then

the system returns an identity such as a username or an ID number.

2.5 LEADING BIOMETRIC TECHNOLOGIES

2.5.1 FINGER-SCAN

Finger-scan is a well-known biometric technology which is used to identify and verify

individuals based on the discriminative features on their fingerprints. Many finger-

scan technologies are based on minutiae points, which are irregularities and

discontinuities characterizing fingerprint ridges and valleys. [3]

2.5.1.1 Advantages of Finger-Scan Technology

It is proven to have very high accuracy.

It does not require complex user – system interaction; therefore little

user training is enough to ensure correct placement of fingers.

It provides the opportunity to enroll up to 10 fingers.

2.5.1.2 Disadvantages of Finger-Scan Technology

High resolution images are required to be acquired due to the small

area of a fingerprint and this results is in more expensive acquisition

devices.

Small percentage of users; elderly populations, manual laborers and

some Asian populations; are shown to be unable to enroll in some

finger-scan systems according to International Biometric Group’s

Comparative Biometric Testing. [3]

As mentioned before, some people may tend to wear down their

fingerprints in time because of their physical work.

Individuals may have objections to collection of their fingerprints

because they may have doubts about usage of their fingerprints for

forensic applications.


PIET Page 14

2.5.2 FACIAL-SCAN

Facial-scan is a biometric technology which is used to identify and verify individuals

based on the discriminative features on their faces. Nonetheless, it is generally used

for identification and surveillance instead of verification. Facial-scan technologies use

some of many discriminative features on face such as eyes, nose, lips etc. [3]

2.5.2.1 Advantages of Facial-Scan Technology

It is the only biometric which provides the opportunity to identify

individuals at a distance avoiding user discomfort about touching a

device.

It can use images captured from various devices from standard video

cameras to CCTV cameras.

2.5.2.2 Disadvantages of Facial-Scan Technology

Changes in lighting conditions, angle of acquisition and background

composition may reduce the system accuracy.

The face is a reasonably changeable physiological characteristic.

Addition or removal of eyeglasses, changes in beard, moustache,

make-up and hairstyle may also reduce the system accuracy.

In order to take changes in environmental conditions and user

appearance into account, facial-scan technologies usually store many

templates for each individual and this results in higher memory

requirement for each individual compared to many other biometrics.

Because face of users may be acquired without their awareness, users

may have objections to facial-scan deployments.

2.5.3 IRIS-SCAN

Iris-scan is a biometric technology which is used to identify and verify individuals

based on the distinctive features on their irises. Iris-scan technologies use the patterns

that constitute the visual component of the iris to discriminate between individuals.[3]


PIET Page 15

2.5.3.1 Advantages of Iris-Scan Technology

It is proven to have smallest FMR among all biometrics, therefore; iris

is the most suitable biometric for applications requiring highest level of

security.

Iris does not change in time, therefore; it does not require reenrollment

which other technologies require after a period of time due to changes

in the biometric.

2.5.3.2 Disadvantages of Iris-Scan Technology

It requires complex user – system interaction, particularly precise

positioning of head and eye. Some systems even require that users do

not move their head during acquisition.

Very high resolution images are required to be acquired due to the

small area of an iris, therefore; acquisition devices are quite expensive.

There is a public objection to using an eye-based biometric even

though many people are not aware of the fact that infrared illumination

is used in iris-scan technology. Were they aware, they might be a much

stronger reaction to this technology.

2.5.4 VOICE-SCAN

Voice-scan is a biometric technology which is used to identify and verify individuals

based on the distinctive aspects of their voice. Voice-scan technologies use different

vocal qualities such as fundamental frequency, short-time spectrum of speech and

spectograms (time – frequency – energy patterns).[3]

2.5.4.1 Advantages of Voice-Scan Technology


PIET Page 16

Various acquisition devices including microphones, land and mobile

phones can be utilized and these devices are relatively cheaper than

acquisition devices used in other biometrics.

Users are prompted to select a pass phrase during enrollment and they

are asked to repeat the same pass phrase during verification and

identification. The probability that imposters guess the correct pass

phrase adds an inherent resistance against false matching.

2.5.4.2 Disadvantages of Voice-Scan Technology

Poor reception quality, ambient noise and echoes may degrade the

system accuracy.

The voice is also a changeable biometric characteristic. Changes in

voice due to illness, lack of sleep and mood may reduce the system

accuracy.

Voice-scan is subject to possibility of recording and replay attacks.

Users are requested to repeat the pass phrase a number of times during

enrollment. Therefore, enrollment process in voice-scan is somewhat

longer than that in other biometrics.

Templates in voice-scan usually occupy a number of times more space

than those in other biometrics.

2.6 DESIRED FEATURES IN A BIOMETRIC

As it is seen, all biometric technologies mentioned above have both advantages and

disadvantages. In other words, there is no perfect biometric technology that has no

disadvantage. However, it is possible to figure out the desired features in a biometric

technology by inspecting advantages and disadvantages of the biometric technologies

above. The list of desired features in a biometric technology is given below:

High Accuracy

Zero or very small FTER

Permanence of biometric in time

Utilization of cheap acquisition devices


PIET Page 17

Resistance to changes in environmental conditions

No or very little public objection (Acceptability)

Small template size

Simple user – system interaction

Inspecting the list above, voice-scan mainly suffers from lower accuracy and higher

template size. Facial-scan may not provide the required accuracy due to changes in

environmental conditions and user appearance. Although iris is the most reliable

biometric, high cost of acquisition devices used in order to scan iris is the biggest

handicap of this technology. Finger-scan has a very high accuracy with simple user

system interaction and small template size. Nevertheless, physical work and age may

cause people not to have clear fingerprints. Additionally, possible dirt and grease on

fingerprints may reduce the system accuracy. Were the area of fingerprint larger,

finger-scan technology might suffer less from effects of dirt, physical work and age

on fingerprints. Palm, on the other hand, provides a large area for feature extraction

and seems to suffer less from factors that reduce the accuracy in finger-scan

technology. Moreover, large area of palm enables utilization of low resolution images

resulting in cheaper acquisition devices. Furthermore, a very small FTER is expected

in palmprint-scan applications because it is easy to correctly place palm on a desired

platform. Due to the same reason, it is possible to have a system with simple user –

system interaction. Additionally, palmprint-scan is a promising biometric technology

to have high accuracy because palmprint is covered with a similar skin as fingerprint.

Finally, palmprint-scan technology has high user acceptance which is quite necessary

for the technology to spread out. As it is seen, palmprint possesses the most of desired

features therefore; it may be used as a biometric. The next chapter will describe some

palmprint recognition algorithms in the literature and will explain results obtained in

these algorithms.


PIET Page 18

CHAPTER 3

EXISTING PALMPRINT RECOGNITION

ALGORITHMS


PIET Page 19

Researchers noticing the increase in biometric revenues in last years and realizing the

advantages of palmprint scan-technology mentioned in the previous chapter started to

develop algorithms to be used in palmprint recognition. Researchers’ interest in

palmprint recognition algorithms has significantly increased especially in last three

years. Due to the fact that the palmprint recognition is a relatively new field of

biometrics, there is a problem related to the utilization of a common palmprint

database in order to be able to compare the performance of different algorithms.

Nevertheless, The Hong Kong Polytechnic University Palmprint Database is the most

commonly used palmprint database. It is here worth giving brief information about

this database before explaining some of the studies on palmprint recognition. The

Hong Kong Polytechnic University Palmprint Database contains 600 grayscale

images corresponding to 100 different palms in Bitmap image format. Palm images

have a resolution of 284x384 pixels with 256 gray levels. Six samples from each of

these palms were collected in two sessions, where 3 samples were captured in the first

session and the other 3 in the second session. The average interval between the first

and the second collection was two months. The palmprint images in the database are

labeled as "PolyU_xx_N.bmp", where the "xx" is the unique palm identifier (ranges

from 00 to 99), and "N" is the index of each palm (ranges from 1 to 6), the palmprints

indexed from 1 to 3 are collected in the first session and 4 to 6 in the second session.

[5] Figure 3.1 shows a schematic diagram of the online palmprint capture device used

to acquire these palm images. The palmprint capture device includes ring source,

CCD camera, lens, frame grabber, and A/D (analogue-todigital) converter. To obtain

a stable palmprint image, a case and a cover are used to form a semi-closed

environment, and the ring source provides uniform lighting conditions during

palmprint image capturing. Also, six pegs on the platform, which is demonstrated in

Figure 3.2, serve as control points for the placement of the user’s hands. The A/D

converter directly transmits the images captured by the CCD camera to a computer.

[6]


PIET Page 20

Figure 3.1:Schematic Diagram of Palmprint Acquisition System [7]

Figure 3.2:Pegs and the Cropped Area of the Palm [7]


PIET Page 21

Various algorithms have been developed to be used in palmprint recognition.

Developed algorithms mainly include different methods for feature extraction and

distance matching. From now on, some of the methods developed for palmprint

recognition will be mentioned and their results will be discussed.

Fang Li et al. [8] proposed an approach utilizing Line Edge Map (LEM) of palmprint

as the feature and Hausdorff distance as the distance matching algorithm. In this

study, Line segment Hausdorff distance (LHD) and Curve segment Hausdorff

distance (CHD) are explored to match two sets of lines and two sets of curves. They

carried out an identification experiment on The Hong Kong Polytechnic University

Palmprint Database. 200 palm images, i.e. 2 palm images for each person, have been

randomly selected in order to test the system performance. They reserved one palm

image for each individual as a template, and used remaining palm images as test

images to be identified. Fang Li et al. [9] later proposed the utilization of Modified

Line segment Hausdorff distance (MLHD) as the distance matching algorithm. In this

study, 2-D lowpass filter is applied to sub-image extracted from the captured hand

image. The result is subtracted from the image in order to decrease the non-uniform

illumination effect resulting from the projection of a 3-D object onto a 2-D image.

After line detection, contour and line segment generation steps, each line on a palm is

represented using several straight line elements. Finally, MLHD is used in order to

measure the similarity between two palm images. Performance of this and some other

palmprint identification methods are tabulated in Table 3-1.

Algorithms employing neural networks have also been developed. Li Shang et al. [13]

suggested the usage of radial basis probabilistic neural network (RBPNN). The

RPBNN is trained by the orthogonal least square algorithm (OLS) and its structure is

optimized by the recursive OLS algorithm (ROLSA). A fast fixed-point algorithm is

used for independent component analysis. The Hong Kong Polytechnic University

Palmprint Database is used to test the developed palmprint recognition algorithm.

After tests performed on this database, recognition rates between % 95 and % 98 are

obtained.


PIET Page 22

CHAPTER 4

PROPOSED SYSTEM


PIET Page 23

Figure 4.1:Block diagram of palm print verification system

A palm print verification system is a one-to-one matching process. It matches a

person’s claimed identity to enrolled pattern. There are two phases in the system:

enrollment and verification. Both phases comprise two sub-modules: preprocessing

for palm print localization, enhancement and feature extraction for moment features

extraction. However, verification phase consists of an additional sub module,

classification, for calculating dissimilarity matching of the palm print. Figure 4 shows

the palm print verification system block diagram.

At the enrollment stage, a set of the template images represented by moment features

is labeled and stored into a database. At the verification stage, an input image is

converted into a set of moment features, and then is matched with the claimant’s palm

print image, based on the ID, stored in the database to gain the dissimilarity measure

by computing Euclidean distance metric. We used this distance metric instead of more

Palm ROI

Template stored in

database

Features

Feature

extraction

Preprocessin

g

Dissimilarity

matching

Features

Palm ROI

Preprocessin

g

Feature

extraction

Threshold

ENROLMENT

IDENTIFICATION


PIET Page 24

complex classification algorithm (e.g. neural network) because we were just focusing

on the feature extracting rather than the classification. Finally, the dissimilarity

measure is compared to a pre-defined threshold to determine whether a claimant

should be accepted. If the dissimilarity measure below the predefined threshold value,

the palm print input is verified possessing same identity as the claimed identity

template and the claimant is accepted.

Also, six pegs on the platform, which is demonstrated in Figure 6 , serve as control

points for the placement of the user’s hands. The A/D converter directly transmits the

images captured by the CCD camera to a computer.

Figure 4.2:Database examples

4.1 ROI EXTRACTION

To extract the region of interest (ROI) from the palm images, the following steps are

to be followed:

Binarization

Contouring

Selecting reference point

Cropping ROI

4.1.1 BINARIZATION

For binarization of the image, we use the global thresholding. Here we find the global

threshold value of the image and compare every pixel of image with the threshold


PIET Page 25

value. If the value is less than the threshold, the pixel value is set to zero; else it is set

to one. For the input image I of the size N×N, global threshold G_Threshold can be

determined using

∑ ∑

……………………………………………...……….(6)

where I(i,j) is intensity value of pixel at position (i,j) of hand image. This threshold is

used to obtain the binarized image BI using

{

………………………………..…(7)

The following figure shows the image of palm and its corresponding binarized image.

This is then further processed using morphological methods for better results.

Figure 4.3:Binarized image

4.1.2 CONTOURING

After getting the binarized image from the palm print image it is then converted to

contour image by using the contour function. The following image shows the

binarized image and its corresponding contour image.


PIET Page 26

Figure 4.4:Palm print contour

4.1.3 SELECTING REFERANCE POINT

Now to select the square or rectangle region on the palm we require a reference point

on the contour. For this we take the distance transform of the contour image. The

distance transform give the distance of the pixel from the nearest non zero value pixel.

From that plot we take the pixel with the highest value, the center most pixel. This

will be the refine pixel to crop ROI. The following figure shows the distance

transform of the palm contour.

Figure 4.5:Distance transform of contour image


PIET Page 27

4.1.4 CROPING ROI

Now the reference point obtained from the distance transform is taken on the palm

print image and from the reference of that point a square or rectangle image is

cropped. The following figure shows the cropped ROI.

Figure 4.6:Region to be cropped

The extracted ROI is then preprocessed and enhanced to make it appropriate for

feature extraction. These are then stored and used for feature matching function for

identification and verification purpose. Figure below shows the square ROI example

that will be used for it.

Figure 4.7:ROI

4.2 ENHANCEMENT OF ROI

The extracted palm print is having non-uniform brightness because of non-uniform

reflection from the relatively curvature of the palm. In order to obtain well distributed

texture image following operations are applied on extracted palm print.


PIET Page 28

(a) (b)

(c) (d)

Figure 4.8:(a) palm print ROI (b) coarse reflection, (c) uniform brightness palm print image, (d)

Enhanced palm print image

The palm print is divided into sub blocks and mean of each sub block is calculated.

Now this image of sub blocks with mean values is subtracted from the original image.

This results in a uniform brightness image. But this is too dark. Now the local

histogram of this image is done to enhance the image.

Palm print is shrinked to the 1/32th

size and zoomed out to 32 times.

This is done with bicubic parameter so as to give estimated coarse

reflection of the image.

This coarse reflection of palm print is then subtracted from the original

ROI to get an uniform brightness image, as shown in figure (c)

The local histogram equalization of this uniform brightness image is done to get

enhanced ROI for further processing and feature extraction.


PIET Page 29

4.3 FEATURE EXTRACTION AND CODING

In this block, relevant features are extracted from the central palm area obtained in the

previous block. Then these extracted features are coded and the mathematical

representation of the palm is obtained. Developing a palmprint recognition algorithm

that can successfully discriminate between palm images of low resolution is a big

advantage from the practical side of view. This is because; since the developed

algorithm does not require high resolution images, there is no need in high resolution

capturing devices which are quite expensive. Being aware of the fact that the cost of

the capturing device plays an important role in determining the total cost of the

developed biometric system, it can be said that the total cost of the system can be

significantly decreased by decreasing the cost of the capturing device. It should be

obvious that low-cost products are easy to market therefore; developing an algorithm

capable of working accurately with low resolution images is very important. Principal

lines, wrinkles, ridges, minutiae points and texture are considered to be relevant

features for a palm (Three principal lines, named as Life Line, Heart Line and Head

Line, and some wrinkles in a palm are shown in Figure 4.14). However, these relevant

features require different resolutions in order to be extracted. In general, principal

lines and wrinkles can be extracted from low resolution images, whereas ridges and

minutiae points need higher resolution. Table 4-1 shows approximate required

resolutions to extract principle lines, wrinkles and ridges texture in dots per inch (dpi).

As it is seen from the table, principal lines can be obtained even in quite low

resolution images. Considering the cost of the biometric system, principal lines may

be thought to be very suitable to be used in the developed algorithm. Although

principal lines can be extracted with algorithms such as the stack filter, they do not

have the uniqueness property, that is, different individuals may have similar principle

lines. This problem has been demonstrated in Figure 4.15. Palm images in (a), (b) and

(c); (d), (e) and (f); and (g), (h) and (i) are very similar to each other; however they

belong to different individuals. Wrinkles may also be thought to be employed,

nevertheless; usage of wrinkles is questionable due to the permanence property,

because wrinkles are subject to change with time. Furthermore, extracting wrinkles

accurately is not an easy task. Due to reasons mentioned above, texture analysis has

been selected to be used in the developed algorithm. [6]


PIET Page 30

Figure 4.9:Principle Lines and Wrinkles in a Palm [20]

Table 4.1: DPI REQUIREMENTS

PALM PRINT FEATURES REQUIRED RESOLUTION (in dpi)

Principal Lines ≥75

Wrinkles ≥100

Ridges Texture ≥125


PIET Page 31

Figure 4.10:Three Sets of Palmprint Images with Similar Principal Lines from Different People

4.4 ZERNIKE MOMENTS

The kernel of Zernike moments is a set of orthogonal Zernike polynomials defined

over the polar coordinate space inside a unit circle. The two dimensional Zernike

moments of order p with repetition q of an image intensity function f(r,θ) are defined

as:

∫ ∫ | |

……………………………….(4.1)


PIET Page 32

Where Zernike polynomials vpq(r,θ) are defined as:

√ …………………………….………………..(4.2)

And the real-valued radial polynomials, Rpq(r), is defined as follows:

∑ | |

( | |

) (

| |

)

………………………………..(4.3)

where 0 ≤ |q| ≤ p and p - |q| is even.

If N is the number of pixels along each axis of the image, then the discrete

approximation of equation (1) is given as:

∑ ∑ ( )

; 0≤ rij ≤1 .....................................(4.4)

where λ(p,N) is normalizing constant and image coordinate transformation to the

interior of the unit circle is given by

√

; (

);

xi = c1 i + c2 ;

yj = c1 j + c2……………………………………………………………………………………………………..……………… (4.5)

Since it is easier to work with real functions, Zpq is often split into its real and

imaginary parts, Zcpq, Z

spq as given below:

∫ ∫

………………………..….(4.6)

∫ ∫

………………………...….(4.7)

where p ≥ 0 , q > 0 .

For the implementation, square image (N x N) is transformed and normalized over a

unit circle; i.e. x2 + y

2 ≤1 , which the transformed unit circle image is bounding the


PIET Page 33

square image. Figure 3 shows the square-to-circular transformation. In this

transformation,

√

√ ………………………………………... (4.8)

Therefore,

√

√ and

√

√ ……………………………………….(4.9)

Figure 4.11:Square to circular transform.

These features are then to be matched with the test image. For that purpose we use

the Euclidean distance. The Euclidean distance between points p and q is the length of

the line line segment ̅̅ ̅.

In Cartesian coordinates, if p = (p1 ,p2,...,pn) and q = (q1 ,q2,...,qn) are two points in

Euclidean n-space, then the distance from p to q is given by

‖ ‖

√


PIET Page 34

The features of the test image and the database are compared using the Euclidean

distance. The image with the least Euclidean distance is considered as the matched

result.


PIET Page 35

CHAPTER 5

RESULTS


PIET Page 36

Experiments were conducted by using a set of database consisting of 20 different

classes of palm prints. Each hand has 10 palm print images. 7 from each are used for

training the system, total 140 images. And other 3 images were used for testing

purpose, total 60 images.

One test image is compared with all the train images to find the corresponding

matching image is. Figure 5.1 to 5.6 shows the minimum distances between the palm

prints of the test image and all train images. The minimum distances are obtained in

the region where the corresponding train images are located. Among them one is

selected as the matched image.

Figure 5.1:Minimum distance for test image 8



PIET Page 37




PIET Page 38




PIET Page 39

Figure 5.7:Value of minimum distance for test image 20

Figure 5.8:Train index values for corresponding test images


PIET Page 40

Figure 5.9:Minimum distance graph for all test images

Figure 5.10:False matched images


PIET Page 41

Figure 5.12:Result of thresholding

Figure 5.12 shows the result of Thresholding. The red colored bars have distances

higher then threshold, thus are eliminated, the green colored bars are truly detected

images. The blue colored bars are false matches, and are less than threshold, thus

giving false matches.

Figure 5.12 displays the histogram of the smallest distance, the distance between the

test images and the most similar templates, for correct matches. Figure 5.13 shows the

histogram of the second smallest distance, the distance between the test images and

the second most similar templates. It is here worth noting that the difference between

the smallest distance and the second smallest distance gives an idea about the

reliability of the identification; that is the bigger the difference is, the more reliable

the identification is. Let the reliability of identification ratio, RI, be defined as the

ratio of this difference to the smallest distance, as in Equation (5.1). The histogram of

the reliability of identification ratio is depicted in Figure 5.14.

RI =

……..…………………………(5.1)


PIET Page 42

Figure 5.13:Smallest distance histogram

Figure 5.14:Second smallest distance histogram


PIET Page 43

Figure 3.15:Reliability of identification

Here database was used and experiment was conducted using different settings of

feature vectors based on the order of ZM and the efficiency is calculated by Euclidean

distance. The efficiency is calculated as the no. of correctly matched images from the

total no. of images. This is then compared to the legendre moments for the same

moment orders. The comparison is shown in Table 5.1

Table 5.1: EFFICIENCY

MOMENT ORDERS ZERNIKE (%) LEGENDRE (%)

0,1 68.3333 73.3333

0,1,2,3 71.6666 66.6666

0,1,2,3,4,5 78.3333 55.0000

0,1,2,3,4,5,6,7 85.0000 66.6666

0,1,2,3,4,5,6,7,8,9 71.6666 33.3333

0,1,2,3,4,5,6,7,8,9,10,11 65.0000 40.0000

The 7th order gives the maximum efficiency of 85%. The other results are then shown

are of this moment order. On the other hand the Legendre moments gives random

change in the efficiency. After 7th

order the efficiency starts reducing due to the noise

affecting the moment calculations.


PIET Page 44

CHAPTER 6

CONCLUSION AND FUTURE SCOPE


PIET Page 45

6.1 CONCLUSION

A palm print identification system using Zernike features is proposed. The proposed

palm print based verification system has the following characteristics:

Constraint free image acquisition: The device used for acquiring hand image

from user should be constraint free. So that physically challenged or injured

people can provide biometric sample.

Robust to translation and rotation: The system should be able to extract palm

print independent to translation and/or rotation of hand on scanner surface.

Low cost scanner: The device used should be economic and easily deployable.

The performance of Zernike moments palm print authentication system was presented

in this thesis. The Zernike moments of order 7 has the best performance among all the

moments. Its efficiency is 85%, which represents the overall performance of this palm

print authentication system. The proposed algorithms, orthogonal moments, possess

some advantages: orthogonality and geometrical invariance. Thus, they are able to

minimize information redundancy as well as increase the discrimination power.

6.2 FUTURE SCOPE

Although performance of the proposed system is satisfactory, it can further be

improved with small modifications and addition preprocessing of hand images.

Also use of circular ROI can be possible by modification in the radial polynomial of

Zernike moments which can make it better rotational invariant.


PIET Page 46

REFERENCES


PIET Page 47

References

PAPERS:

[1] John D. Woodward, Jr., Nicholas M. Orlans, Peter T. Higgins, “Biometrics”,

McGraw-Hill, 2003.

[2] Ruud M. Bolle, Jonathan H. Connell, Sharath Pankanti, Nalini K. Ratha, Andrew

W. Senior, “Guide to Biometrics”, Springer, 2004.

[3] Samir Nanavati, Michael Thieme, Raj Nanavati, “Biometrics”, Wiley, 2002.

[4] “Biometrics Glossary”, retrieved 10 January 2007, from

http://www.biometricscatalog.org/biometrics/GlossaryDec2005.pdf

[5] “PolyU Palmprint Palmprint Database”

http://www.comp.polyu.edu.hk/~biometrics/

[6] David Zhang, Wai-Kin Kong, Jane You, Michael Wong, “Online Palmprint

Identification”, IEEE Transactions on Pattern Analysis and Machine Intelligence,

Vol. 25, No. 9, pp 1041-1050, September 2003.

[7] M. Wong, D. Zhang, W.-K. Kong and G. Lu, “Real-time Palmprint Acquisition

System Design”, IEE Proc.-Vis. Image Signal Process., Vol. 152, No. 5, October

2005.

[8] Fang Li, Maylor K.H. Leung, Xiaozhou You, “Palmprint Identification Using

Hausdorff Distance”, 2004 IEEE International Workshop on Biomedical Circuits &

Systems, 2004.

74

[9] Fang Li, Maylor K.H. Leung, Xiaozhou You, “Palmprint Matching Using Line

Features”, ICACT 2006, 20-22 February 2006.

[10] N. Duta, A.K. Jain, “Matching of Palmprints”, Pattern Recognition Letters I,

Vol. 23, pp 477-485, 2002.

[11] J. You, W. X. Li and D. Zhang, “Hierarchical Palmprint Identification via

Mutiple Feature Extraction”, Pattern Recognition Vol. 35, No. 4, pp 847-859, 2002.

[12] D. Zhang, W. Shu, “Two Novel Characteristics in Palmprint Verification:

Datum Point Invariance and Line Feature Matching”, Pattern Recognition, vol 33,

no. 4, pp 691-702, 1999.

[13] Li Shang, De-Shuang Huang, Ji-Xiang Du, Chun-Hou Zheng, “Palmprint

Recognition using FastICA algorithm and radial basis probabilistic neural network”,

2005.


PIET Page 48

[14] Xiang-Qian Wu, Kuan-Quan Wang, David Zhang, “Palmprint Recognition

Using Valley Features”, Proceedings of the Fourth International Conference on

Machine Learning and Cybernetics, Guangzhou, 18-21 August 2005._

[15] A.K. Jain, S. Prabhakar, L. Hong, S. Pankanti, “Filterbank-Based Fingerprint

Matching”, IEEE Trans. Image Process., Vol. 9, No. 5, pp 846–859, 2000.

[16] L. Hong, Y. Wan, A. Jain, “Fingerprint Image Enhancement Algorithm and

Performance Evaluation”, IEEE Trans. Pattern Anal. Mach. Intell., Vol. 20, No. 8, pp

777–789, 1998.

[17] C.J. Lee, S.D. Wang, “Fingerprint Feature Extraction Using Gabor Filters”,

Electron. Lett., Vol. 35, No. 4, pp 288–290, 1999.

75

[18] Xiangqian Wu, Kuanquan Wang, Fengmiao Zhang, David Zhang, “Fusion of

Phase and Orientation Information for Palmprint Authentication”, 2005.

[19] Ajay Kumar, Helen C. Shen, “Palmprint Identification using PalmCodes”,

Proceedings of the Third International Conference on Image and Graphics, 2004.

[20] Wai King Kong, David Zhang, Wenxin Li, “Palmprint Feature Extraction Using

2-D Gabor Filters”, Pattern Recognition, Vol. 36, pp 2339- 2347, 2003.

[21] M.J. Lyons, J. Budynek, S. Akamatsu, “Automatic Classification of Single

Facial Images”, IEEE Trans. Pattern Anal. Mach. Intell., Vol. 21, No. 12, pp 1357–

1362, 1999.

[22] B. Duc, S. Fischer, J. Bigun, “Face Authentication with Gabor Information on

Deformable Graphs”, IEEE Trans. Image Process., Vol. 8, No. 4, pp 504–516, 1999.

[23] Y. Adini, Y. Moses, S. Ullman, “Face Recognition: The Problem of

Compensation for Changes in Illumination Direction”, IEEE Trans. Pattern Anal.

Mach. Intell., Vol. 19, No. 7, pp 721–732, 1997.

[1] Ying-Han Pang, Andrew T.B.J, David N.C.L, Hiew Fu San “Palm print

Verification with Moments” Journal of WSCG, Vol.12, No.1-3, ISSN 1213-

6972 WSCG’2004, February 2-6

[2] Amir Tahmasbi, FatemehSaki,ShahriarB.Shokouhi “Classification of benign

and malignant masses based on Zernike moments” Elsevier- Computers in

Biology and Medicine 2011, june 14

[3] Madasu Hanmandlu, Neha Mittal, Ankit Gureja, Ritu Vijay “A

Comprehensive Study of Palmprint based Authentication” International

Journal of Computer Applications (0975 – 8887) vol. 37 – No.2, January 2012

[4] Atif Bin Mansor, Hassan Masood, Mustaffa Mumtaz, Shoab A. Khan “ A

feature level multimodal approach for palmprint identification using


PIET Page 49

directional subband energies” Elsevier – Journal of network and Computer

Applications.

BOOKS:

[1] Flusser, Jan, Suk, Tomáš and Zitová, Barbara. “Moments and Moment

Invariants.”

[2] Liao, Simon Xin meng. “Image Analysis by moments.”

WEBSITES:

[1] http://en.wikipedia.org/wiki/Moment_(mathematics)

[2] http://www4.comp.polyu.edu.hk/~biometrics/index_db.htm

http://www.si2.org/openeda.si2.org/dfmcdictionary/index.php/Zernike_Pol

ynomials

http://en.wikipedia.org/wiki/Moment_(mathematics)

http://www4.comp.polyu.edu.hk/~biometrics/index_db.htm

http://www.si2.org/openeda.si2.org/dfmcdictionary/index.php/Zernike_Polynomials

http://www.si2.org/openeda.si2.org/dfmcdictionary/index.php/Zernike_Polynomials


PIET Page 50

APPENDIX


PIET Page 51

APPENDIX-A

sharma thesis1

Documents