CHAPTER 3 STUDY OF EXISTING UNIMODAL BIOMETRIC...

CHAPTER – 3

STUDY OF EXISTING UNIMODAL BIOMETRIC SYSTEMS AND DESIGNING

AND DEVELOPMENT OF VARIOUS UNIMODAL BIOMETRIC SYSTEMS

3. 1 Unimodal Biometric Techniques

Biometric techniques can be divided into two categories based on no. of traits which are

considered to decide identity of a person. Biometric techniques which are using single

traits for identification or verification of person are known as Unimodal Biometric

Techniques. Biometric techniques which are using multiple algorithms, multiple traits,

multiple sensors or multiple samples are known as Multibiometric Techniques.

Biometric systems are categorized in two categories based on traits used for person

identification. Physiological biometric systems judge the person based on physical

characteristics of human being. Behavioral biometric systems judge the person based on

behavioral characteristics of human being. Example physiological biometric systems are:

1. Fingerprint recognition

2. Face recognition

3. Iris recognition

4. Retina scanning

5. Hand geometry

6. Palmprint recognition

Example behavioral biometric systems are:

1. Voice recognition

2. Gait recognition

3. Keystroke recognition

In this research, we are concentrating on fingerprint and face recognition techniques, so

in this chapter, we will take review of only physiological biometrics systems. We have

studied fingerprint and face recognition systems in detail in chapter 2, so here we will

take review of other unimodal biometric systems like Iris recognition, Retinal scanning,

hand geometry etc. Other physiological biometric systems are not so popular in

implementation. So, we will restrict overview with above mentioned three technologies.

3.1.1 Iris recognition

Iris recognition is one of the most accurate biometric systems now a day. By considering

this feature, iris recognition has been successfully implemented in ATMs and kiosks for

banking and travel applications.

Iris recognition system contains front end acquisition hardware along with local or central

processing software. Compared to facial recognition, iris recognition required specialized

devices which provide infrared illumination.

The software components of iris system – image processing and matching systems,

database will reside on local PC along with attached device or can reside on central

system. In large scale system, central server will do the work of matching templates and

storing database. The local system will do the work of acquiring sample and generating

template. The central device and local PC will communicate image template instead of

image itself.

Based on the results of matching physical or logical access to the resources is granted.

3.1.1.1 History of iris recognition [1]

1936: US ophthalmologist Frank Burch suggests the idea of recognizing people

from their iris patterns long before technology for doing so is feasible.

1981: American ophthalmologists Leonard Flom and Aran Safir discuss the idea

of using iris recognition as a form of biometric security, though technology is still

not yet advanced enough.

1987: Leonard Flom and Aran Safir gain US patent #4,641,349 for the basic

concept of an iris recognition system.

1994: US-born mathematician John Daugman (currently a professor of computer

science at Cambridge University, England) works with Flom and Safir to develop

the algorithms (mathematical processes) that can turn photographs of irises into

unique numeric codes. He is granted US patent #5,291,560 for a "biometric

personal identification system based on iris analysis" the same year. Daugman is

widely credited as the inventor of practical iris recognition since his algorithm is

used in most iris-scanning systems.

1996: Lancaster County Prison, Pennsylvania begins testing iris recognition as a

way of checking prisoner identities.

1999: Bank United Corporation of Houston, Texas converts supermarket ATMs

to iris-recognition technology.

2000: Charlotte/Douglas International Airport in North Carolina and Flughafen

Frankfurt Airport in Germany become two of the first airports to use iris scanning

in routine passenger checks.

2006: Iris-scanning systems are installed at British airports, including Heathrow,

Gatwick, Birmingham, and Stansted. Privacy concerns notwithstanding, hundreds

of thousands of travelers voluntarily opt to use the machines to avoid lengthy

passport-checking queues.

3.1.1.2 Uniqueness of iris recognition

The iris is the colored ring of muscle that opens and shuts the pupil of the eye like a

camera shutter. The colored pattern of our irises is determined genetically when we're in

the womb but not fully formed until we're aged about two. It comes from a pigment

called melanin—more melanin gives you browner eyes and less produces bluer eyes. The

color and pattern of people's eyes is extremely complex and completely unique: the

patterns of one person's two eyes are quite different from each other and even genetically

identical twins have different iris patterns.

Figure 3.1: Iris image

8

3.1.1.3 Working of iris recognition

To get past an iris-scanning system, the unique pattern of your eye has to be recognized

so you can be positively identified. That means there have to be two distinct stages

involved in iris-scanning: enrollment (the first time you use the system, when it learns to

recognize you) and verification/recognition (where you're checked on subsequent

occasions).

Enrollment

First, all the people the system need to know about have to have their eyes scanned. This

one-off process is called enrollment. Each person stands in front of a camera and has their

eyes digitally photographed with both ordinary light and invisible infrared (a type of light

used in night vision systems that has a slightly longer wavelength than ordinary red

light). In iris recognition, infrared helps to show up the unique features of darkly colored

eyes that do not stand out clearly in ordinary light. These two digital photographs are then

analyzed by a computer that removes unnecessary details (such as eyelashes) and

identifies around 240 unique features (about five times more "points of comparison" as

fingerprint systems use). These features, unique to every eye, are turned into a simple,

512-digit number called an IrisCode® that is stored, alongside your name and other

details, in a computer database. The enrollment process is completely automatic and

usually takes no more than a couple of minutes. [1]

Figure 3.2: Iris scanners

Verification

Once you're stored in the system, it's a simple matter to check your identify. You simply

stand in front of another iris scanner and have your eye photographed again. The system

quickly processes the image and extracts your IrisCode®, before comparing it against the

hundreds, thousands, or millions stored in its database. If your code matches one of the

stored ones, you're positively identified; if not, tough luck! It either means you're not

known to the system or you're not whom you claim to be. [1]

3.1.1.4 Strengths and weaknesses of iris recognition

Strengths of iris recognition:

1. It has the potential for exceptionally high levels of accuracy.

2. It is capable of reliable identification as well as verification.

3. It maintains stability of characteristic over a lifetime.

Weaknesses of iris recognition:

1. Acquisition of the image requires moderate training and attentiveness.

2. It has a propensity for false rejection.

3. A proprietary acquisition device is necessary for deployment.

4. There is some user discomfort with eye-based technology

3.1.1.5 Applications of iris recognition

The accuracy of the technology appeals to high-security applications such as military and

national infrastructure; and its remote-acquisition capability and ease of use lend

themselves to high-throughput technology and screening applications such as airports and

checkpoints.

The United Arab Emirates uses iris-recognition technology to screen all incoming visitors

against a list of thousands of persons who have been expelled from the UAE. Border

authorities have done 200 billion cross-comparisons between IrisCodes and have caught

46,000 persons illegally attempting to reenter the UAE — with no false matches.

In the United States, the Child Project is an iris-based system for helping to identify and

return missing children; as of September 2007, 1,400 sheriff‘s offices were participating.

The company that supplies the technology for the Child Project (BI2 Technologies) also

supplies Senior Safety Net and the Inmate Recognition and Identification System (IRIS).

The U.S. is using iris-recognition technologies in Iraq to control access to facilities, but

has so far resisted the temptation to do more than capture facial images on passports.

3.1.2 Retina scanning

Retina-scan technology utilizes the distinctive characteristic of the retina—the surface on

the back of the eye that processes light entering through the pupil— for identification and

verification. Developed in the 1980s, retina-scan is one of the most well-known biometric

technologies, but is also one of the least deployed.

Retina-scan devices are used exclusively for physical access applications and are usually

used in environments requiring exceptionally high degrees of security and accountability

such as high-level government, military, and corrections applications.

Retina-scan and iris-scan are often mistakenly confused with one another or grouped into

a single category referred to as eye biometrics. The two technologies differ substantially.

They measure different physiological features, the software and algorithm technology is

very different, iris- and retina-scan hardware and software are dissimilar, and the

situations in which they can be successfully deployed differ.

3.1.2.1 History of retina scan

Retina biometrics distinguishes individuals by using the patterns of veins occurring in the

back of the eye. A 1935 study by Drs. C. Simon and I. Goldstein first observed the

individually distinguishing characteristics of retinal vascular patterns. Automated

techniques to capture and process retinal patterns for recognition were developed in the

1970s along with the first wave of other early pioneering efforts in digital imaging.

Established in 1976, EyeDentify of Baton Rouge, Louisiana, made retinal scanning

commercially available for access control in the early 1980s [13].

3.1.2.2 Uniqueness of retina scan

The retina‘s intricate network of blood vessels is a physiological characteristic that

remains stable throughout the life of a person. As with fingerprints and iris patterns,

genetic factors do not determine the exact pattern of blood vessels in the retina. This

allows retina-scan to differentiate between identical twins and provide robust

identification. The retina contains at least as much individual data as a fingerprint, but,

unlike a fingerprint, is an internal organ and is less susceptible to either intentional or

unintentional modification. Certain eye-related medical conditions and diseases, such as

cataracts and glaucoma, can render a person unable to use retina-scan technology, as the

blood vessels can be obscured [2].

3.1.2.3 Working of retina scan

The retina scan works with image acquisition, identifying distinctive features and

template generation and matching [4].

Image acquisition

Since the retina is small, internal, and difficult to measure without the proprietary

hardware and camera systems specifically designed for retina imaging, image acquisition

is a very difficult process.

Figure 3.4: Retina scanner

In order for the unit to acquire retina images, the user first positions his or her eye very

close to the unit‘s embedded lens, with the eye socket resting on the sight. Beneath the

lens, within the device itself is an imaging component consisting of a small green light

Figure 3.3: Retina

against a white background. The user views this light through the lens; when triggered,

the light moves in a tight circle, measuring the retinal patterns through the pupil. In order

for a retinal image to be acquired, the user must gaze directly into the lens, remaining

perfectly still while focusing on the imaging component. Any movement defeats the

image acquisition process and requires that the imaging process be triggered again. A

small camera captures an image of the retina through the pupil. The acquisition of a

single retina image takes 4 to 5 seconds under ideal conditions. During enrollment,

between three and five acceptable images must be acquired. Since the first one or two

images acquired are almost invariably rejected due to excessive movement, the

enrollment process can be relatively lengthy. Including the time to trigger the process,

respond to system prompts, and acquire sufficient images, enrollments can easily take

over 1 minute. Many users cannot enroll at all, even after several minutes. On the other

hand, it is possible for highly acclimated users to be identified within 2 to 3 seconds—the

identification process is much quicker [4].

Distinctive feature

The retina‘s intricate network of blood vessels is distinctive feature. Even identical twins

have different patterns of blood vessels [4].

Template generation and matching

Once a device captures a retinal image, the software compiles the unique features of the

network of retinal blood vessels into a template. Retina-scan algorithms require a high-

quality image and will not let a user enroll or verify until the system is able to capture an

image of sufficient quality. The retina template generated by the originator of the

technology is a mere 96 bytes, one of the smallest of any biometric technology.

Retina-scan has robust matching capabilities and is typically configured to do one-to-

many identification against a database of users. However, because quality image

acquisition is so difficult, many attempts are often required to get to the point where a

match can take place. While the algorithms themselves are robust, it can be a difficult

process to provide sufficient data for matching to take place. In many cases, a user may

be falsely rejected because of an inability to provide adequate data to generate a match

template [4].

3.1.2.4 Strengths and weaknesses of retina scan

Strengths include the following:

1. It is highly accurate.

2. It uses a stable physiological trait.

3. It is very difficult to spoof.

Weaknesses include the following:

1. It is very difficult to use.

2. There is some user discomfort with eye-related technology.

3. It has limited applications.

3.1.2.5 Applications of retina scan

Retinal scan devices are mainly used for physical access applications and are usually

used in environments requiring exceptionally high degrees of security and accountability

such as high-level government, military, and corrections applications. Retinal scanning

has been utilized by several U.S. government agencies including the Federal Bureau of

Investigation (FBI), the Central Intelligence Agency (CIA), and NASA.

Retinal scanning is also used for medical diagnostic applications. Examining the eyes

using retinal scanning can aid in diagnosing chronic health conditions such as congestive

heart failure and atherosclerosis.

3.1.3 Hand geometry

Today, the human hand has another use, a media to verify identity. Ancient Egyptians

used body measurements to classify and identify people. Today‘s hand geometry

scanners use infrared optics and microprocessor technology to quickly and accurately

record and compare hand dimensions [5].

3.1.3.1 History of Hand geometry

Several hand geometry verification technologies have evolved during this century. They

range from electro-mechanical devices to the solid state electronic scanners being

manufactured today. The U.S. Patent office issued patents to Robert P. Miller in the late

1960‘s and early 1970‘s for a device that measures hand characteristics, and records

unique features for comparison and ID verification. Miller‘s machines were highly

mechanical and manufactured under the name ―Identimation.‖ Several other companies

launched development and manufacturing efforts during the 70‘s and early 80‘s. In the

mid- 1980‘s, David Sidlauskas developed and patented an electronic hand scanning

device and established the Recognition Systems, Inc. of Campbell, California in 1986.

The first applications for hand scanners were as access control components. Government

and nuclear facilities used them to protect their facilities [5].

3.1.3.2 Working of hand geometry

Today, hand scanners perform a variety of functions including access control, employee

time recording and point-of sale applications.

Figure 3.5: Hand geometry scanner

Each human hand is unique. Finger length, width, thickness, curvatures and relative

location of these features distinguish every human being from every other person. The

hand geometry scanner uses a charge coupled device (CCD) camera, infrared light

emitting diodes (LEDs) with mirrors and reflectors to capture black and white images of

the human hand silhouetted against a thirty-two thousand pixel field. The scanner records

no surface details, ignoring fingerprints, lines, scars and color. The process is much like

placing a hand on a beaded projector screen. The hand scanner reads the hand shape by

recording the silhouette of the hand. In combination with a side mirror and reflector, the

optics produces two distinct images, one from the top and one from the side. This method

is known as orthographic scanning [5].

Figure 3.6: Measurement of different parameters

Scanners typically use an optical path approximately 11 inches between the camera and

the platen. An optical path folded with mirrors reduces the space required to half the

original length. Enclosing the optical path in a structure results in the typical hand

geometry scanner that is approximately 8-1/2 inches square by 10‖ high. The scanner

takes ninety-six measurements of the user‘s hand. A microprocessor and internal software

convert the measurements to a nine-byte ‗‘template‖ that it stores for later comparison.

The process of recording a user‘s hand template is known as enrollment. During the

enrollment session, the scanner prompts the enrollee to place his or her hand on the

scanner platen three consecutive times. The platen is the highly reflective surface that

projects the silhouetted hand image. Pins projecting from the platen surface position the

enrollee‘s fingers to assure accurate image capture. The hand geometry scanner

mathematically averages the three templates and generates an accurate template that the

scanner stores in resident memory. To verify, the user enters a personal identification

number (PIN) in the scanner through the use of a keypad or other data entry device. The

scanner retrieves his or her individual template for comparison. The user places his or her

hand on the scanner. The hand image is captured and a representation is derived using the

same steps as those used for generating the template at the time of enrollment. The

representation thus derived is compared to the stored template. The comparison may

involve, for instance, accumulation of absolute differences in the individual features in

the input representation and the stored template. The comparison typically results in a

single number indicating the strength of the similarity (score) or their difference

(distance). A predetermined threshold determines whether the score/distance is

acceptable to consider the input representation and stored templates are ―matched‖. The

match/no-match decision controls the output of the scanner.

3.1.3.3 Implementation issues of hand geometry

Hand geometry technology faces some implementation issues to be considered, which are

given as below:

1. Card reader emulation

2. Stand alone access control

3. Privacy issues

4. Operation by disabled person

5. Outdoor conditions

3.1.3.4 Strengths and weaknesses of hand geometry

Strengths include the following:

1. Ease of use

2. Resistant to fraud

3. Template size

4. User perception

Weaknesses include the following:

1. Static design

2. Cost

3. Injuries to hands

4. Accuracy

3.1.3.5 Applications of hand geometry

Hand geometry biometric systems have variety of applications. The domains are many

where it can be used. The list of areas is given below:

1. Parking lot application

2. Cash vault application

3. Dual custody application

4. Anti-passback

5. Time and attendance

6. Point of sale applications

7. Interactive kiosks

3.1.4 Limitations of unimodal biometric systems

3.1.4.1 Reasons for failure of different unimodal biometric systems

Fingerprint

1. Cold finger

2. Dry/oily finger

3. High or low humidity

4. Angle of placement

5. Pressure of placement

6. Location of finger on platen (poorly placed core)

7. Cuts to fingerprint

8. Manual activity that would mar or affect fingerprints (construction, gardening)

Facial recognition

1. Change in facial hair

2. Change in hairstyle

3. Lighting conditions

4. Adding/removing hat

5. Adding/removing glasses

6. Change in weight

7. Change in facial aspect (angle at which facial image is captured)

8. Too much or too little movement

9. Quality of capture device

10. Change between enrollment and verification cameras (quality and placement)

11. ‗Loud‘ clothing that can distract face location

Iris-scan

1. Too much movement of head or eye

2. Glasses

3. Colored contacts

Retina-scan

1. Too much movement of head or eye

2. Glasses

Hand geometry

1. Jewelry

2. Change in weight

3. Bandages

4. Swelling of joints

3.1.4.2 Limitations of unimodal biometric systems

While unimodal biometric systems have advantages over password or token based

approaches, they have several challenges also. Here are some challenges of unimodal

biometric systems [6]:

1. Noise in the sensed data

A fingerprint image with a scar, or a voice sample altered by cold are examples of

noisy data. Noisy data may also result from defective or improperly maintained

sensors or unfavorable ambient conditions. Noisy biometric data may not be

successfully matched with corresponding templates in the database, resulting in a

genuine user being incorrectly rejected.

2. Intra-class variations

Intra-class variations in biometric systems are typically caused by an individual who

is incorrectly interacting with the sensor (e.g., incorrect facial pose), or due to

changes in the biometric characteristics of a person over a period of time (e.g., change

in hand geometry). These variations can be handled by storing multiple templates for

every user and updating these templates over time. Template update is an essential

ingredient of any biometric system since it accounts for changes in a person's

biometric with the passage of time. The face, hand and voice modalities, in particular,

can benefit from suitably implemented template update mechanisms.

3. Inter-class similarities

Inter-class similarity refers to the overlap of feature spaces corresponding to multiple

classes or individuals. In an identification system comprising of a large number of

enrolled individuals, the interclass similarity between individuals will increase the

false match rate of the system. Therefore, there is an upper bound on the number of

individuals that can be effectively discriminated by the biometric system.

4. Non-universality

The biometric system may not be able to acquire meaningful biometric data from a

subset of users. A fingerprint biometric system, for example, may extract incorrect

minutia features from the fingerprints of certain individuals, due to the poor quality of

the ridges. Thus, there is a failure to enroll (FTE) rate associated with using a single

biometric trait.

5. Interoperability issues

Most biometric systems operate under the assumption that the biometric data to be

compared are obtained using the same sensor and, hence, are restricted in their ability

to match or compare biometric data originating from different sensors. For example, a

speaker recognition system may find it challenging to compare voice prints

originating from two different handset technologies such as electret and carbon-

button.

6. Spoof attacks

Spoofing involves the deliberate manipulation of one's biometric traits in order to

avoid recognition or the creation of physical biometric artifacts in order to take on the

identity of another person. This type of attack is especially relevant when behavioral

traits such as signature and voice are used. However, physical traits such as

fingerprints and iris are also susceptible to spoof attacks. Spoof attacks, when

successful, can severely undermine the security afforded by a biometric system.

There are several ways to address issues related to spoofing. In the case of physical

traits, such as fingerprint and iris, a liveness detection scheme may be used to detect

artifacts; in the case of behavioral traits, a challenge-response mechanism may be

employed to detect spoofing.

7. Other vulnerabilities

A biometric system is vulnerable to a broad range of attacks. Ratha et al., 2001

identified several levels of attacks that can be launched against a biometric system: (i)

a fake biometric trait such as an artificial finger may be presented at the sensor, (ii)

illegally intercepted biometric data may be resubmitted to the system, (iii) the feature

extractor may be replaced by a Trojan horse program that produces pre-determined

feature sets, (iv) legitimate feature sets may be replaced with synthetic feature sets,

(v) the matcher may be replaced by a Trojan horse program that always outputs high

scores thereby defying system security, (vi) the templates stored in the database may

be modified or removed, or new templates may be introduced in the database, (vii)

the data in the communication channel between two modules of the system may be

altered, and (viii) the final decision output by the biometric system may be

overridden.

3.2 Creation of database for face and fingerprint recognition system

In the first phase of research, the researcher prepared two unimodal systems viz. face and

fingerprint recognition system. As a part of development of both unimodal systems, first

step is to create master database for both unimodal systems. Here is the directory

structure for storage of image database of 30 persons. The researcher has considered 30

persons for preparation of face and fingerprint database. Here the researcher has

considered 25 students and 5 faculty members. First 25 persons are students of

Department of Computer Science, Saurashtra University, Rajkot. Last 5 persons are

faculty members and the researcher himself. The details of persons are shown below:

Sr. No. Name of student / faculty Gender Folder name Index of the person

1 Chandarana Niyati F s1 101

2 Vaishnav Kairavi F s2 102

3 Kalariya Monali F s3 103

4 Dave Malvika F s4 104

5 Kambariya Jigna F s5 105

6 Shekhda Ravina F s6 106

7 Bhagiya Divya F s7 107

8 Rathod Riddhi F s8 108

9 Kalariya Manali F s9 109

10 Gohel Ruchika F s10 110

11 Padiya Seema F s11 111

12 Kavar Dipali F s12 112

13 Maru Arjun M s13 113

14 Chavda Gaurav M s14 114

15 Bhalodiya Pratik M s15 115

16 Dodiya Chirag M s16 116

17 Shah Parth M s17 117

18 Gosai Kaushikgiri M s18 118

19 Makwana Nayan M s19 119

20 Kapadiya Dhaval M s20 120

21 Shah Moin M s21 121

22 Dangar Jaydev m s22 122

23 Chirag Gusani M s23 123

24 Vyas Abhay M s24 124

25 Chavda Ravi M s25 125

26 Divyakant Meva M s26 126

27 Apurva Pandya M s27 127

28 Shital Rakangor F s28 128

29 Dr C K Kumbharana M s29 129

30 Hetal Thaker F s30 130

Table 3.1: Person details of face and fingerprint database with directory name structure

Before describing the model of building database, here is the description of devices used

for the said purpose.

3.2.1 Device used for capturing face samples

The researcher has used Dell 1525 Inspiron system‘s inbuilt camera for capturing facial

images of different persons. The camera has been manufactured by Creative

Technologies Ltd. The camera captures image of the size 320x240 pixels. Then image is

converted into gray scale image. 320x240 pixels images are then resized with 112x92

pixels. The image is saved in .pgm format as it occupies less size. The figure of the

device is shown in figure 3.7.

3.2.2 Device used for capturing fingerprint samples

The researcher has used Digital Persona U.are.U 4000 scanner. The specifications are

shown below in figure 3.8. The device captures image in .jpg format with resolution of

328x356. The device is shown in figure 3.9.

Figure 3.7: Dell webcam used for capturing face samples

Figure 3.8: Digital Persona U.are.U 4000 / 4000S / 4000B specifications

Figure 3.9: Digital Persona U.are.U 4000 used for capturing fingerprint samples

3.2.3 Directory structure for storage of captured face and fingerprint samples

In this research, 30 persons have been considered for database creation. The details are

shown in table 3.1. For each person, 10 samples are captured for face and fingerprint

each. The face samples are taken with various postures as well as based on presence of

spectacles. Few samples are captures with expressions. The reason is to identify whether

system is capable of identifying person even if he has variations in spectacles and

expressions. For each person total 20 images have been captured. Total 600 images –

samples are there in database.

Total number of persons: 30

Number of faces per person: 10

Number of fingerprints per person: 10

Total number of face and fingerprint images per person: 10 + 10= 20

Total no. of images in face database: 10 * 30 = 300

Total no. of images in fingerprint database: 10 * 30 = 300

Total no. of images in face and fingerprint database: 300 + 300 = 600

The directory structure is shown and explained here.

Here there are two directories viz. Face database and Fingerprint database. Under these

directories, 30 subdirectories are there in each directory. The structure for face database

directory is shown in figure 3.10. The naming convention is here: name of the directory

starts with ‗s‘ and is clubbed with the number 1 to 30 based on index of person. The

example directory name is ‗s1‘ for the first person. And will continue up to ‗s30‘ for

thirtieth person. The structure of sub directory is shown in figure 3.11. Every

subdirectory indeed contains 10 samples. The name of each image is generated in the

following manner. The name of sample starts with index number of person i.e. ‗101‘ and

followed by ‗_‘ and sample number starting from 1 to 10. Example structure for face

recognition database is given below:

Face recognition s1101_1.pgm

The sample image file structure is shown in figure 3.12.

Similarly for fingerprint recognition database structure is given below:

Fingerprint recognition s1 101_1.jpg

The sample image file structure is shown in figure 3.13, 3.14 and 3.15.

Figure 3.10 Base directory structures for face database

Figure 3.11: Sub directory structure for face database directory

Figure 3.12: File structure for ‗s1‘ sub directory in ‗Face database‘ directory

Figure 3.13 Base directory structures for fingerprint database

Figure 3.14: Sub directory structure for fingerprint database directory

Figure 3.15: File structure for ‗s1‘ sub directory in ‗Fingerprint database‘ directory

3.2.4 Model for database creation

The researcher has prepared GUI with which database preparation will be easier. The

device details are shown above. The model for preparing database is shown below:

Figure 3.16: Model for creation of master databases

The model works in the following manner:

Initially, the user requires executing IDE for capturing fingerprint and face. Then by

pressing button, first webcam will be initiated and face sample will be captured. After

capturing face sample, demo.exe file from ZKFinger SDK will be executed to capture

fingerprint image and both the images will be renamed as per the requirement of user and

will be stored in respective directories created automatically by system. After creating

base database of face and fingerprint images, next is to prepare master train database of

face and fingerprint images. The process of creating master train database is mentioned in

section 3.2.5.

The master database which is created at the end of process contains 60 face sample (2

samples of each person – 30x 2) and 30 fingerprint samples (1 sample of each person

30x1).

The steps to be followed to create database with GUI are shown below:

Step-1: Load IDE for capturing face and fingerprint samples. Click on the button

„Capture face and fingerprint‟

Main engine to

create database

Database of 60

face and 30

fingerprint

templates

Figure 3.17: Load IDE for capturing face and fingerprint samples

Execute capturefaceandfinger.m for execution of IDE to capture face and fingerprint

samples. The IDE contains two axes controls and two edit box controls, three buttons.

Axes controls shows face and fingerprint images. Then click on the button ‗Capture face

and fingerprint‘.

Step-2: Dell webcam will capture face sample and will load in IDE

On clicking ‗Capture face and fingerprint‘, Dell webcam will be initiated and will capture

face sample and will save that sample image as ‗temp.pgm‘ in root folder. The face

sample then will be loaded in axes control1. The GUI is shown in figure 3.18.

Step-3: demo.exe from ZKFinger SDK will be loaded to capture

After loading face sample, demo.exe file of ZKFinger SDK will be executed. The GUI of

demo.exe is shown in Figure 3.19. Click ‗Connect Sensor‘ button to initiate fingerprint

sensor. ZKFinger SDK is the one, through which it possible to build various fingerprint

recognition application. Digital Persona‘s U.are.U 4000 sensor can be connected and

used to capture fingerprint samples with demo.exe of ZKFinger SDK.

Axes control 1 Axes control 2

Edit control 1 Edit control 2

Figure 3.18: IDE loaded with face sample captured by Dell webcam

Figure 3.19: Execution of demo.exe of ZKFinger IDE

Step-4: Select image format and ZKFinger 9.0 version

Figure 3.20: Captured fingerprint sample image in demo.exe GUI

After loading demo.exe GUI and connecting sensor, select ZKFinger 9.0 and Image

format ‗.jpg‘. On putting finger on U.are.U 4000 fingerprint sensor, it will be visible in

demo.exe GUI.

Step-5: Click button „Save Image‟ and close demo.exe

After selecting appropriate options, click on ‗Save Image‘ button on demo.exe GUI to

save captured fingerprint sample. The fingerprint sample will be saved as

‗fingerprint.jpg‘ in root folder. The GUI is shown in figure 3.21. Then close demo.exe

GUI. Control will be passed to IDE and fingerprint sample will be loaded on axes

control2 of IDE.

Step-6: Enter person index in editbox 1 and sample index in editbox 2

In the next step, enter person index in edit box1 and sample index in edit box2. As per the

directory structure, person index will start from 101 and sample index will start from 1.

Figure 3.21: On exiting demo.exe, fingerprint sample loaded in IDE

Step-7: Click on „Rename files‟ button

Figure 3.22: Click on ‗Rename files‘ button

After entering person index and sample index, click Rename files‘ button. On clicking

this button, face.pgm and fingerprint.jpg files from root folder will renamed as the

filename generated by combining person index and sample index, i.e. 131 as person index

and 1 as sample index then ‗131_1.pgm‘ for face.pgm and ‗131_1.jpg‘ for

fingerprint.jpg. Once renaming files, sample index will automatically change to 2 in edit

box2. After completing 10 samples for one person, person index will automatically

increase to next index i.e. from ‗131‘ to ‗132‘. Both the files will automatically move to

the folder with person index name. Figure 3.22 and 3.23 shows this representation.

Figure 3.23: Sample index changes automatically and files will be renamed

Step-8: Click on „Clear images‟ button

After completing the operations of capturing face and fingerprint samples and renaming

files and moving them to respective folders, click on ‗Clear images‘ button to clear the

content of axes controls. Figure 3.24 shows this representation of operation.

After completing sample capturing operations for face and fingerprint for all persons,

prepare directory structure shown in 3.10, 3.11, 3.13 and 3.14. Move all face and

fingerprint images in their respective directories and subdirectories. For example,

samples of person index ‗101‘ will be moved to ‗s1‘ subdirectory of Face recognition and

Fingerprint recognition directories. The final file structure is shown in 3.12 and 3.15.

Figure 3.24: Click ‗Clear images‘ button to clear content of axes control

3.2.5 Identify train and test samples for master database creation

Once completing preparation of sample images database, next step is to identify sample

images for master database, which are also called train samples. This step is shown as

step-3 in database creation model in figure 3.16. Identifying train samples for face master

database, the researcher has adopted novel approach, which has been discussed hereafter.

For identifying train samples for master fingerprint database, the researcher identified

good quality fingerprint sample images manually. And db.mat is created by extracting

minutiae features from train fingerprint sample images and creating templates, which are

stored in .mat file with respective person index number. The structure of fingerprint

database files is shown like: person index e.g. 101 which is followed by ‗_‘ and again

followed by ‗1‘.

e.g. 101_1 and so on, up to 101_30

3.2.5.1 Identifying train face samples for master database

The initial database is created by taking 30 objects with each enrolled with 10 samples of

face. Total no. of samples is 300. The problem here is how to identify samples for

training database and testing database. We planned to keep 2 samples with training

database and 8 samples in testing database for each person.

Figure 3.25: PCA based face recognition model

Initially two sample images of each person are included in train database. As we have 30

persons, 60 images would be there in train database. Features are extracted from images

by applying Eigenface method. Out of 300 images, rest 240 images are added in test

database. Now, by applying Euclidean distance method for Eigenface approach,

minimum distance will be calculated. The image with minimum distance from train

database sample with test sample will be identified. And success rate will be counted. In

this system, we have taken five rounds of the above discussed flow of process. Initially,

sample image 1 and 2 of each person were placed in train database and rest of the images

(8 images) was placed in test database. In the second round, images 3 and 4 were placed

in train database and compared with rest test samples to identify success rate. Round

three contained sample 5 and 6, round 4 contained samples 7 and 8 and round five

contained samples 9 and 10 as train database. The results of this approach to find best

images suitable to add in train database are shown below.

Results of experiment

Considering a database of 60 sample facial images in Train Database of 30 persons (2 for

each person) and taking database of 240 sample facial images in Test Database of 30

persons (8 for each person)

No. of persons = 30

No. of images in Train Database = 60 (Two for each person)

No. of images in Test Database = 240 (Eight for each person)

Case 1 – Image 1, 2 as train database and rest as test database





Case

number

Image

number Success rate in %

Samples identified successfully

(Out of 240)

1 1,2 63.75 153

2 3,4 57.50 138

3 5,6 74.58 179

4 7,8 70.41 169

5 9,10 62.08 149

Table 3.2: Success rate of five cases

Based on the results shown in table 3.2, the researcher considered sample 5 and 6 as train

dataset and rest of the samples as test dataset.

By considering the above shown operations, 60 face samples have been identified as

train database samples and 30 fingerprint samples have been identified to build db.mat

master database file by extracting minutiae features.

3.3 Building unimodal face recognition system

Once preparing database of face and fingerprint samples, and identifying samples for

train database and test database, the next step is to prepare unimodal face and fingerprint

recognition systems. First, the researcher has designed unimodal face recognition model

to identify person based on face features. Here the researcher has used PCA based face

recognition method using Eigenface approach. The original system has been developed

by Amir Hossein Omidvarnia. He has prepared the system by taking the reference from

[12]. The model of face recognition system is shown below:

Figure 3.26: Face recognition system model

Figure 3.26 shows the model of face recognition system. Initially multimodal IDE is

executed and face recognition system option is selected. In that GUI, click the button to

capture face sample. The process of capturing face sample is similar to that mentioned in

section 3.2.4. After that, the sample image will be compared master train database of 60

images (2 images for each person * 30 person). By comparing Euclidean distance

between master database images and sample image, the image with minimum Euclidean

distance will be identified and index of that image will be returned and displayed in GUI.

3.3.1 Flow chart representation of face recognition system

The system operations can be described with the flow chart shown in figure 3.27. The

Unimodal and Multimodal IDE have been developed to execute the operations of face

recognition and fingerprint recognition.

Figure 3.27: Flowchart of unimodal face recognition system

Execute Unimodal IDE

and select Face

recognition option

Load test face

sample from

test database

Master face

database

Calculate Euclidean

distance between test

sample and train sample

from master face database

Display person

index with

minimum

Euclidean

distance

End process

Start

Identify the image with

minimum Euclidean

Distance

For ease of use to the user, the researcher has developed this IDE. The steps are shown in

flow chart and GUI execution is shown after the flow chart representation. As per the

system design, the user can select test image from database. It is possible to implement

the mechanism of enrolling live sample for face recognition. It is possible to evaluate the

performance without using IDE i.e. with the help of command prompt execution of .m

file.

Based on the flowchart shown in figure 3.27, here are the GUI steps for face recognition

system execution.

3.3.2 GUI representation of face recognition system

Step-1: Load multimodal IDE

Figure 3.28: Multimodal IDE with all components

The first step is to load multimodal IDE. For that, it is required to execute

multimodal_ide.m file. The GUI contains three panels containing different components.

Axes control 1 &

Axes control 2

Edit controls

Menu selection

options

The panels are: controls, Test images, Scores and Identity. Based on selection of menu

option, components will be visible on IDE.

Step-2: Select „Unimodal‟ menu option from two menus and from Unimodal menu select

„Face recognition‟ option.

IDE contains two menu options: ‗Unimodal‘ and ‗Multimodal‘. As here is the face

recognition system execution, so ‗Unimodal‘ option requires to be selected. On selecting

this option, two options - ‗Face recognition‘ and ‗Fingerprint recognition‘ will be visible.

Figure 3.29 shows these options.

Step-3: Loading components required for face recognition system

On selection of Unimodal menu option and Face recognition submenu option, the

components required for face recognition system will only be visible to the user. The

components are: Load face image button, Match face button and Clear image button, axes

control 1 to load test face image, and edit control to display recognized person index.

Figure 3.29: Selection of Unimodal menu option and Face recognition submenu option.

Figure 3.30: Components for face recognition system

Step-4: Selection of test face image from test database

Once loading components for face recognition system, next step is to press ‗Load face

image‘ button. On pressing this button, ‗Pick an image file‘ dialog box will be visible and

select a test face image with extension .pgm from test databases.

Step-5: Display selected test face image in axes control1 and press „Match face‟ button

After selection of test face image from test database, the system will load this file on axes

control 1. Next, press ‗Match face‘ button to perform face recognition process. The

system will execute PCA based face recognition method using Eigenface approach. The

working of method is shown in section 3.3.2.1.

Figure 3.31: Selection of test face image

3.3.2.1 PCA based face recognition system with Eigenfaces [8][9][12]

Principal component analysis transforms a set of data obtained from possibly correlated

variables into a set of values of uncorrelated variables called principal components. The

number of components can be less than or equal to the number of original variables. The

first principal component has the highest possible variance, and each of the succeeding

components has the highest possible variance under the restriction that it has to be

orthogonal to the previous component. It is required to find the principal components;

here eigenvectors of the covariance matrix of facial images.

(1) First step

In the first step, it is required to form a training data set. 2D image can be represented in

1D vector with concatenation of rows. Image will be transformed in a vector with length

N=m*n.

I=

x11 x12 … x1n

x21 x22 … x2n

⋮ ⋮ ⋱ ⋮xm1 xm2 … xmn

x11

⋮x1n

⋮x2n

⋮xmn

=x

Create a matrix of learning images X with M vectors of length N. Then center the matrix.

Determine vector of mean values and subtract that vector from each image vector.

Average vectors are arranged to form a new training vector with size (NxM).

(2) Second step

Second step will calculate covariance matrix C, and find its eigenvectors and eigenvalues.

Covariance vector C will have dimension NxN. From this matrix, we can get N

eigenvectors and eigenvalues. Rank of covariance matrix is limited by number of images

in learning set. Eigenvector associated with highest eigenvalue reflects the highest

variance and one associated with the lowest eigenvalue, the smallest variance.

The vectors should be sorted by eigenvalues so that first vector corresponds to the highest

eigenvalue. These vectors will be normalized next. This will create a new matrix where

each vector is a column vector. The dimension of this matrix will be NxD, where D

represents desired number of eigenvectors. Each original image can be reconstructed by

adding mean image to weighted sum of all vectors.

(3) Third step

Third and last step is the recognition of faces. Image of the person required to find in

training database is transformed into a vector P, reduced by the mean value and projected

with a matrix of eigenvectors (Eigenfaces).

Classification is done by identifying distance person‘s eigenvector and each vector of

matrix Y. Euclidean distance is the most common method. Other methods can also be

applied.

If the minimum distance between test face and training faces is higher than threshold

value, then person will be unknown otherwise it will be known one.

mxn

CONCATENATION

1xn

Step-6: Displaying identity of person in edit control

Figure 3.32: Display identity of person with person index in edit control

As shown in step-5, system will execute PCA based face recognition method using

Eigenface approach and identify the index of person having minimum Euclidean distance

with test face sample. After this identification, next step is to click on ‗Clear all‘ button to

clear the content of axes control and edit control.

3.3.3 Performance evaluation of face recognition system

For performance measurement of this unimodal biometric technique, the researcher has

taken three cases.

1. By considering average distance of 8 test samples

2. By considering minimum distance of 8 test samples

3. By considering maximum distance of 8 test samples

Every train database sample is compared with 60 train database samples and calculated

Euclidean distance. This process is done for all 8 test database samples and from that

minimum, maximum and average Euclidean distance is calculated and stored in 3

different files with names: min_distance.txt, max_distance.txt and avg_distance.txt. After

writing all these scores in text files, it is required to calculate False Accept Rate (FAR)

and False Reject Rate (FRR) for the test samples. For FRR, test is applied with all 8 test

samples and compared with minimum score, maximum score and average score stored in

respective text files. Based on that, False Reject Rate (FRR) is calculated. The outcomes

of this experiment are given below:

Calculating FRR

Consider a database of 60 samples for training and 240 samples for testing. Here training

set for each person with 2 samples and other 8 samples of the same person compared by

taking different cases.

Case 1: minimum distance from all 8 test samples with 2 train samples

Case 2: maximum distance from all 8 test samples with 2 train samples

Case 3: average distance of all 8 test samples with 2 train samples

The Matlab code for calculating FRR is shown here:

-----------------------------------------------------------------------

% A sample script, which shows the usage of functions, included in % PCA-based face recognition system (Eigenface method) % This program calculates average minimum distance for each person by % considering image 5 and 6 as train sample and rest 8 as test samples.

clear all clc close all

% You can customize and fix initial directory paths TrainDatabase =

'D:\phd_dtm_practical\FaceRecognition_University\Experiment3\Database\

TrainDatabase'; TestDatabase =

'D:\phd_dtm_practical\FaceRecognition_University\Experiment 3\Database\

TestDatabase'; TestDatabase2 =

'D:\phd_dtm_practical\FaceRecognition_University\Experiment3\TestDataba

se';

count=100;

success_min=0; success_max=0; success_avg=0;

frr_min=0;

frr_max=0; frr_avg=0;

% Fetching min_distance value from file min_distance = []; max_distance = []; avg_distance = [];

fid=fopen('min_distance.txt','r+'); fid1=fopen('max_distance.txt','r+'); fid2=fopen('avg_distance.txt','r+'); fid3=fopen('falserejection.txt','wt'); % Generate vector of minimum distance for i=1:30 distance1=fscanf(fid,'%f'); min_distance = [min_distance distance1]; end

% Generate vector of maximum distance for i=1:30 distance1=fscanf(fid1,'%f'); max_distance = [max_distance distance1]; end

% Generate vector of average distance for i=1:30 distance1=fscanf(fid2,'%f'); avg_distance = [avg_distance distance1]; end

% Finding FRR for each person based on minimun distance for i=1:30 TrainDatabasepath=strcat(TrainDatabase,int2str(i)); T = CreateDatabase(TrainDatabasepath,i); [m, A, Eigenfaces] = EigenfaceCore(T); TestDatabasepath = strcat(TestDatabase,int2str(i)); for j = 1 : 8 num=count+i; str=strcat(int2str(num),'_',int2str(j)); TestImage = strcat(TestDatabasepath,'\',str,'.pgm'); im = imread(TestImage); min_dist = Recognition(TestImage, m, A, Eigenfaces); %fprintf([num2str(min_dist) '\t' num2str(min_distance(i))

'\n']); if(min_dist>min_distance(i)) frr_min=frr_min+1; else success_min=success_min+1; end end end

disp(frr_min); disp(success_min);

fprintf(fid3,'%f \t %f',frr_min,success_min);

% Finding FRR for each person based on maximum distance for i=1:30 TrainDatabasepath=strcat(TrainDatabase,int2str(i)); T = CreateDatabase(TrainDatabasepath,i); [m, A, Eigenfaces] = EigenfaceCore(T); TestDatabasepath = strcat(TestDatabase,int2str(i)); for j = 1 : 8 num=count+i; str=strcat(int2str(num),'_',int2str(j)); TestImage = strcat(TestDatabasepath,'\',str,'.pgm'); im = imread(TestImage); min_dist = Recognition(TestImage, m, A, Eigenfaces); %fprintf([num2str(min_dist) '\t' num2str(min_distance(i))

'\n']);

if(min_dist>max_distance(i)) frr_max=frr_max+1; else success_max=success_max+1; end end end

disp(frr_max); disp(success_max);

fprintf(fid3,'%f \t %f',frr_max,success_max);

% Finding FRR for each person based on average distance for i=1:30 TrainDatabasepath=strcat(TrainDatabase,int2str(i)); T = CreateDatabase(TrainDatabasepath,i); [m, A, Eigenfaces] = EigenfaceCore(T); TestDatabasepath = strcat(TestDatabase,int2str(i)); for j = 1 : 8 num=count+i; str=strcat(int2str(num),'_',int2str(j)); TestImage = strcat(TestDatabasepath,'\',str,'.pgm'); im = imread(TestImage); min_dist = Recognition(TestImage, m, A, Eigenfaces); %fprintf([num2str(min_dist) '\t' num2str(min_distance(i))

'\n']);

if(min_dist>avg_distance(i)) frr_avg=frr_avg+1; else success_avg=success_avg+1; end end end

disp(frr_avg); disp(success_avg);

fprintf(fid3,'%f \t %f',frr_avg,success_avg);

fclose(fid); fclose(fid1); fclose(fid2); fclose(fid3); -----------------------------------------------------------------------

Results of calculating FRR

The results of experiment of calculation of FRR are shown in table 3.3 and graphical

representation is shown in figure 3.33.

Case No. of train

samples

No. of

test

samples

Genuine

acceptance

False

rejection GAR FRR

min_dist 60 240 26 214 10.83% 89.16%

max_dist 60 240 142 98 59.16% 40.83%

avg_dist 60 240 240 0 100 0

Table 3.3: FRR calculation for three different cases

Figure 3.33: FRR calculation for facial recognition

Calculating FAR

Consider a database of 60 samples for training. Here training set for each person with 2

samples and other 290 samples of the other persons compared by taking different cases.

0

20

40

60

80

100

120

GAR FRR

min_dist

max_dist

avg_dist

Case 1: minimum distance from all 8 test samples with 2 train samples

Case 2: maximum distance from all 8 test samples with 2 train samples

Case 3: average distance of all 8 test samples with 2 train samples

The understanding of comparisons is shown below:

For each person 10 face images are there, so for 30 persons, here 300 images are there.

As we have to calculate FRR, so comparison of facial image of a person will be made

with rest 29 persons‘ images. i.e. facial image of person 101 will be compared with facial

images of person 102 to 130. So, comparison of facial image of person 101 will be there

with 290 other images. The similar comparison is carried out for rest 29 persons. So, total

8700 (290 * 30 persons) comparisons will be made.

The Matlab code for calculating FAR is shown here:

-----------------------------------------------------------------------

% A sample script, which shows the usage of functions, included in PCA-

based face recognition system (Eigenface method) % This program calculates FAR by comparing two samples of one person

with % 290 samples of other 29 persons

clear all clc close all

% You can customize and fix initial directory paths TrainDatabase =

'D:\phd_dtm_practical\FaceRecognition_University\Experiment

3\Database\TrainDatabase'; TestDatabase =

'D:\phd_dtm_practical\FaceRecognition_University\Experiment

3\TestDatabase\TestDatabase';

count=100;

success_min=0; success_max=0; success_avg=0;

far_min=0; far_max=0; far_avg=0;

% Fetching min_distance value from file min_distance = []; max_distance = []; avg_distance = [];

fid=fopen('min_distance.txt','r+'); fid1=fopen('max_distance.txt','r+');

fid2=fopen('avg_distance.txt','r+'); fid3=fopen('falseaccept.txt','wt'); % Generate vector of minimum distance for i=1:30 distance1=fscanf(fid,'%f'); min_distance = [min_distance distance1]; end

% Generate vector of maximum distance for i=1:30 distance1=fscanf(fid1,'%f'); max_distance = [max_distance distance1]; end

% Generate vector of average distance for i=1:30 distance1=fscanf(fid2,'%f'); avg_distance = [avg_distance distance1]; end

% Finding FAR for each person based on minimun distance for i=1:30 TrainDatabasepath=strcat(TrainDatabase,int2str(i)); T = CreateDatabase(TrainDatabasepath,i); [m, A, Eigenfaces] = EigenfaceCore(T); for j = 1 : 30 TestDatabasepath = strcat(TestDatabase,int2str(j)); for l=1:10 fprintf(['value of i is:' num2str(i) ' value of j is:'

num2str(j) 'value of l is:' num2str(l) '\n']); if(i==j) continue; else num=count+j; str=strcat(int2str(num),'_',int2str(l)); TestImage = strcat(TestDatabasepath,'\',str,'.pgm'); im = imread(TestImage); min_dist = Recognition(TestImage, m, A, Eigenfaces); if(min_dist<min_distance(i)) far_min=far_min+1; else success_min=success_min+1; end end end end end disp(far_min); disp(success_min);

fprintf(fid3,'%f \t %f \n',far_min,success_min);

% Finding FAR for each person based on maximum distance for i=1:30

TrainDatabasepath=strcat(TrainDatabase,int2str(i)); T = CreateDatabase(TrainDatabasepath,i); [m, A, Eigenfaces] = EigenfaceCore(T); for j = 1 : 30 TestDatabasepath = strcat(TestDatabase,int2str(j)); for l=1:10 if(i==j) continue; else num=count+j; str=strcat(int2str(num),'_',int2str(l)); TestImage = strcat(TestDatabasepath,'\',str,'.pgm'); im = imread(TestImage); min_dist = Recognition(TestImage, m, A, Eigenfaces); if(min_dist<max_distance(i)) far_max=far_max+1; else success_max=success_max+1; end end end end end disp(far_max); disp(success_max);

fprintf(fid3,'%f \t %f \n',far_max,success_max);

% Finding FAR for each person based on average distance for i=1:30 TrainDatabasepath=strcat(TrainDatabase,int2str(i)); T = CreateDatabase(TrainDatabasepath,i); [m, A, Eigenfaces] = EigenfaceCore(T); for j = 1 : 30 TestDatabasepath = strcat(TestDatabase,int2str(j)); for l=1:10 if(i==j) continue; else num=count+j; str=strcat(int2str(num),'_',int2str(l)); TestImage = strcat(TestDatabasepath,'\',str,'.pgm'); im = imread(TestImage); min_dist = Recognition(TestImage, m, A, Eigenfaces); if(min_dist<avg_distance(i)) far_avg=far_avg+1; else success_avg=success_avg+1; end end end end end disp(far_avg); disp(success_avg);

fprintf(fid3,'%f \t %f \n',far_avg,success_avg);

fclose(fid); fclose(fid1); fclose(fid2); fclose(fid3); -----------------------------------------------------------------------

Results of calculating FAR

The results of experiment of calculation of FRR are shown in table 3.4 and graphical

representation is shown in figure 3.34.

Case False acceptance

(Out of 8700)

Genuine

Rejection

(Out of 8700)

FAR GRR

min_dist 153 8547 1.75% 98.25%

max_dist 6727 1973 77.32% 22.67%

avg_dist 2981 5791 34.26% 66.56%

Table 3.4: FAR calculation for three different cases

Figure 3.34: FAR calculation for facial recognition

The performance of the face recognition system has been shown in table 3.3 and table

3.4. The details discussion and comparison of performance with other system‘s

performance will be there in chapter 6.

0

20

40

60

80

100

120

FAR GRR

min_dist

max_dist

avg_dist

3.4 Building unimodal fingerprint recognition system

After preparing database of face and fingerprint samples, and identifying samples for

train database and test database, the next step is to prepare unimodal face and fingerprint

recognition systems. The design of unimodal face recognition model to identify person

based on face features is discussed in section 3.3. Here is the description of unimodal

fingerprint recognition system. The researcher has used minutiae based fingerprint

recognition approach. The original system has been developed by Vahid. K. Alilou of

Department of Computer Engineering from The University of Semnan.. The model of

fingerprint recognition system is shown below in figure 3.35.

Initially multimodal IDE is executed and fingerprint recognition system option is

selected. In that GUI, click the button to capture fingerprint sample. The process of

capturing fingerprint sample is similar to that mentioned in section 3.2.4. After that, the

sample image will be compared master train database (db.mat) of 30 images (1 image for

each person * 30 person). By comparing minutiae point features between master database

images and sample image, the image with maximum score will be identified and index of

that image will be returned and displayed in GUI.

Figure 3.35: Fingerprint recognition system model

3.4.1 Flow chart representation of fingerprint recognition system

The system operations can be described with the flow chart shown in figure 3.36. The

steps are shown in flow chart and GUI execution is shown after the flow chart

representation. As per the system design, the user can select test image from database. It

is possible to implement the mechanism of enrolling live sample for fingerprint

recognition. The IDE provides facility for both face and fingerprint recognition under the

same menu option i.e. ‗Unimodal‘.

Figure 3.36: Flowchart of unimodal fingerprint recognition system

Execute Unimodal IDE

and select fingerprint

recognition option

Load test

fingerprint

sample from

test database

Master fingerprint

database db.mat

Identify minutiae

features from test

samples and compare

with db.mat templates

Maximum

matching

score?

Display person

index with

maximum

matching score

End process

Start

The details of steps are shown below after representation of flow chart. Based on the

flowchart shown in figure 3.36, here are the GUI steps for fingerprint recognition system

execution.

3.4.2 GUI representation of fingerprint recognition system execution

Step-1: Load multimodal IDE

The first step is to load multimodal IDE. For that, it is required to execute

multimodal_ide.m file. The GUI contains three panels containing different components.

The panels are: controls, Test images, Scores and Identity. Based on selection of menu

option, components will be visible on IDE. See the figure 3.37.

Step-2: Select „Unimodal‟ menu option from two menus and from Unimodal menu select

„Fingerprint recognition‟ option.

IDE contains two menu options: ‗Unimodal‘ and ‗Multimodal‘. As here is the face

recognition system execution, so ‗Unimodal‘ option requires to be selected. On selecting

this option, two options - ‗Face recognition‘ and ‗Fingerprint recognition‘ will be visible.

Figure 3.38 shows these options.

Figure 3.37: Multimodal IDE with all components

Figure 3.38: Selection of Unimodal option and Fingerprint recognition submenu option

Step-3: Loading components required for fingerprint recognition system

On selection of Unimodal menu option and Fingerprint recognition submenu option, the

components required for fingerprint recognition system will only be visible to the user.

The components are: Load fingerprint image button, Match fingerprint button and Clear

image button, axes control 2 to load test fingerprint image, and edit control to display

recognized person index. See the figure 3.39.

Step-4: Selection of test fingerprint image from test database

Once loading components for fingerprint recognition system, next step is to press ‗Load

fingerprint image‘ button. On pressing this button, ‗Pick an image file‘ dialog box will be

visible and select a test fingerprint image with extension .jpg from test databases. See

figure 3.40 for detail.

Step-5: Display selected test fingerprint image in axes control2 and press „Match

fingerprint‟ button

After selecting test fingerprint image from test database, the system will load this file on

axes control2. Next, press ‗Match fingerprint‘ button to perform fingerprint recognition

process. See figure 3.41 for GUI representation. The system will execute minutiae based

fingerprint recognition approach. The working of method is shown in 3.4.3.

Figure 3.39: Components for fingerprint recognition system

Fingerprint matching algorithms largely follow 3 different classes: correlation-based,

minutiae-based, and non-minutiae feature based matching. Correlation-based matching

involves superimposing 2 fingerprint images together and calculating pixel-wise

correlation for different displacement and rotations. Minutia-based matching uses

extracted minutiae from both fingerprints in order to help perform alignment and retrieve

minutiae pairings between both fingerprint minutiae sets. Minutiae-based matching can

be viewed as a point-pattern matching problem with theoretical roots in pattern

recognition and computer vision. Non-minutiae feature based matching use non-minutiae

features, such as ridge shape, orientation and frequency images in order to perform

alignment and matching. Amongst all algorithm classes, minutiae-based methods are the

most common due to their strict analogy with the way forensic experts compare

fingerprints and legal acceptance as a proof of identity in many countries. Minutiae points

are also known to be extremely unique from finger to finger in terms of spatial

distribution, proving to be ideal features for fingerprint matching. Additionally, minutiae

point sets obtain a higher level of uniqueness versus practicality in comparison to other

level types of fingerprint features, such as ridge orientation/frequency images and skin

pores. Here with this research, the researcher has adopted minutiae based fingerprint

recognition method.

Figure 3.40: Selection of test fingerprint sample image from test database

3.4.2.1 Minutiae based fingerprint recognition system [10][11]

The researcher has adopted the work of Vahid Alilou from University of Semnan, Iran.

The open source code has been provided by him. He has developed minutiae based

fingerprint recognition system. With minor changes, the researcher has adopted the code.

The graphical representation of the system steps is given below:

The simple explanation is given below:

The complete processing is done in three stages:

1. Preprocessing

2. Feature extraction

3. Matching

Step 1 comprises the following steps:

a. Thinning

b. Binarization Making

c. segmentation mask

d. Image enhancement

Figure 3.41: Loading test fingerprint image in axes control2


a. Finding minutiae points

b. Filtering false minutiae points


a. Loading database

b. Registration

c. Computing matching score

Figure 3.42: Fingerprint recognition system steps

Stage 2 contains the following mechanism:

The algorithm, first, extracts some features from the fingerprint and stores them in a

vector called 'minutiae' which contains the following data: [X, Y, CN, Theta, Flag, 1]

where X, and Y contains the coordination of the a minutiae, Theta is the orientation of the

minutiae. CN is the Crossing Number . {0: Isolated point, 1: Termination Point, 2:

Continuing Ridge Point, 3: Bifurcation Point} .Flag is 0 for permissible minutiae and is 1

for non-permissible one.

The function 'transform', transforms x, y, theta according to the i-th reference point. The

function 'match', finds the best matching using minutiae feature details and computes the

overall similarity score.

Step-6: Displaying identity of person in edit control

Figure 3.43: Display identity of the recognized person with edit control

As shown in step-5, system will execute minutiae based fingerprint matching method to

identify best matching score and based on that the person index. The recognized person‘s

index will be shown in edit control. After this identification process, next is to click on

‗Click all‘ button to clear the content of axes control and edit control.

3.4.3 Performance evaluation of fingerprint recognition system

The researcher carried out experiment of fingerprint recognition system by taking 9 test

samples of each person, in these way 270 samples of 30 persons. These samples are

compared with 30 train database samples. This experiment returns GAR and FRR for the

fingerprint recognition system.

Calculating FRR


set for each person with 1 sample and other 9 samples of the same person compared by

taking different cases. The Matlab code for calculating FRR is shown here:

-----------------------------------------------------------------------

clear all; clc; addpath(genpath(pwd)); load('db.mat'); GAR=0; FAR=0; fid=fopen('fp_far1.txt','wt'); count=100; for i=9:30 person=count+i; for j=1:9

filename=['D:\phd_dtm_practical\FingerprintRecognition_University\TestD

atabase\' num2str(person) '_' num2str(j) '.jpg']; img=imread(filename); if ndims(img) == 3; img = rgb2gray(img); end % Color Images disp(['Extracting features from ' filename ' ...']); ffnew=ext_finger(img,1); fprintf(['Computing similarity between ' num2str(j) ' and '

'database file 101_1 ' 'from SU2014 :']); str1= [person '_' num2str(j) '.jpg']; score=match(ffnew,ff{i}); if (score >0.50) GAR=GAR+1; else FRR=FRR+1; end fprintf(fid,'%s %f \n',str1,score); end end display(GAR); display(FRR); fprint(fid,'%f \t %f \n',GAR,FRR); fclose(fid);

-----------------------------------------------------------------------

By applying the above given code, the researcher has achieved the following results

shown in table 3.5. Graphical representation is shown in 3.44.

Case Success

(Out of 270)

Failure

(Out of 270) GAR FRR

1 257 13 95.185% 4.815%

Table 3.5: GAR and FRR for fingerprint recognition system

The researcher carried out experiment of fingerprint recognition system by taking 9 test

samples of each person, in these way 270 samples of 30 persons. These samples are

compared with 30 train database samples. This experiment returns GAR and FRR for the

fingerprint recognition system.

Figure 3.44: GAR and FRR for fingerprint recognition system

Calculating FAR

The researcher carried out another experiment of fingerprint recognition system by taking

90 test samples of 30 persons. These samples are compared with 30 train database

samples. This experiment returns GRR and FAR for the fingerprint recognition system.


set for each person with 1 sample and other 3 samples of the different persons compared.

The Matlab code for calculating FAR is shown here:

-----------------------------------------------------------------------

clear all; clc; addpath(genpath(pwd)); load('db.mat'); GRR=0; FAR=0; fid=fopen('fp_far.txt','wt'); count=100; for i=1:30 str=[ 'comparison with' count+i '.jpg' ]; fprintf(fid,'%s \n',str); for j=1:3 person=count+i+j; if person>130 person=person-30; end

filename=['D:\phd_dtm_practical\FingerprintRecognition_University\TestD

atabase\' num2str(person) '_' num2str(j) '.jpg']; img=imread(filename); if ndims(img) == 3; img = rgb2gray(img); end % Color Images disp(['Extracting features from ' filename ' ...']); ffnew=ext_finger(img,1);

0

20

40

60

80

100

GAR FRR

Series1

fprintf(['Computing similarity between ' num2str(j) ' and '

'database file' num2str(i) 'from SU2014 :']); str1= [person '_' num2str(j) '.jpg']; score=match(ffnew,ff{i}); if (score >0.48) FAR=FAR+1; else GRR=GRR+1; end fprintf(fid,'%s %f \n',str1,score); end end display(GRR); display(FAR); fprint(fid,'%d \t %d \n',GRR,FAR); fclose(fid); -----------------------------------------------------------------------

By applying the above given code, the researcher has achieved the following results

shown in table 3.6. Graphical representation is shown in 3.45.

Table 3.6: GRR and FAR for fingerprint recognition system

Figure 3.45: GRR and FAR for fingerprint recognition system

From the experiments, we are sure about performance of fingerprint recognition system.

The success rate (GAR) of fingerprint recognition system is 95.18%. Similarly it has

acceptable FAR. But these results are in standard conditions. If there are some problems

like oily skin, scars and dirt on the finger skin, the results will be poorer.

0

20

40

60

80

100

GRR FAR

Series1

Case Success

(Out of 90)

Failure

(Out of 90) GRR FAR

1 88 2 97.77% 2.23%

3.5 Conclusion of prototyping unimodal biometric systems

From the experiments carried out with facial and fingerprint recognition system, we are

able to conclude the following results.

Sr. no. Biometric system type GAR FRR GRR FAR

1 Face recognition (min_dist) 10.83% 89.16% 1.75% 98.25%

2 Face recognition (max_dist) 59.16% 40.83% 77.32% 22.67%

3 Face recognition (avg_dist) 100% 0% 34.26% 66.56%

4 Fingerprint recognition 95.185% 4.815% 97.77% 2.23%

Table 3.7: Performance comparison of facial and fingerprint recognition systems

With these experiments, we can say that fingerprint recognition system can give

acceptable performance under standard conditions. But in case of noise like oily skins,

scars, dirt and in case of injury, it may not be able to give acceptable performance. At the

same time, facial recognition system alone is not capable to give acceptable performance.

In this situation, we can say that we require considering more than one modality instead

of only one biometric trait or unimodal biometric system. To get optimum performance

of the system, we should use multimodal system with more than one modality.

References

[1] www.explainthatstuff.com

[2] www.biometricupdate.com

[3]Peter Gregory, Michael Simon, ―Biometrics for Dummies‖, Wiley publishing, Inc,

2008

[4]Anil Jain, Ruud Bole, Sharath Pankanti, ―Biometrics – Personal identification in

networked society‖, Kluwer academic publishers, 2002

[5]Samir Nanavati, Michael Thieme, Raj Nanavati, ―Biometrics – Identity verification in

a networked world‖, Wiley computer publishing, 2002

[6] Arun Ross, Karthik Nandkumar, Anil Jain, ―Handbook of Multibiometrics‖, Springer,

2006

[7] Ratha, N. K., Connell, J. H., and Bolle, R. M., ―An Analysis of Minutiae Matching

Strength‖, In Proceedings of Third International Conference on Audio- and Video-Based

Biometric Person Authentication (AVBPA), pages 223-228, Halmstad, Sweden., 2001

http://www.explainthatstuff.com/

http://www.biometricupdate.com/

[8] M. Turk, A. Pentland: Face Recognition using Eigenfaces, Conference on Computer

Vision and Pattern Recognition, 3 – 6 June 1991, Maui, HI , USA, pp. 586 – 591.

[9] Marijeta Slavković, Dubravka Jevtić, ‖Face recognition using Eigenface approach‖,

Serbial Journal of Electrical Engineering, Vol. 9, no. 1, 2012, pp. 121-130

[10] Joshua Abraham, Paul Kwan and Junbin Gao, ―Fingerprint Matching using A

Hybrid Shape and Orientation Descriptor‖, State of the art in Biometrics, Dr. Jucheng

Yang (Ed.), ISBN: 978-953-307-489-4, InTech, 2011.

[11] Vahid K. Alilou, Fingerprint matching – A simple approach, www.mathworks.com

[12] P. N. Belhumeur, J. Hespanha, and D. J. Kriegman. Eigenfaces vs. Fisherfaces:

Recognition using class specific linear projection. In ECCV (1), pages 45--58, 1996.

[13] John Woodword, Nicholas Orlans, Peter Higgins,‖ Biometrics : The ultimate

reference‖, Wiley India Publications

http://www.mathworks.com/

CHAPTER 3 STUDY OF EXISTING UNIMODAL BIOMETRIC...

Documents

Transcript of CHAPTER 3 STUDY OF EXISTING UNIMODAL BIOMETRIC...