IRIS SEGMENTATION PROCESS - csnow.in SEGMENTATION PROCESS ... Kattamanchi Hemachandran, ... I would...

IRIS SEGMENTATION PROCESS

A PROJECT SUBMITTED TO ASSAM UNIVERSITY, SILCHAR IN PARTIAL

FULFILLMENT OF THE REQUIREMENT FOR THE DEGREE OF THE

MASTERS OF COMPUTER SCIENCE

SUBMITTED BY

SAHENA BEGAM BARBHUIYA

Msc 4th sem

Roll : No : 22220384

Regn. No: 01-110016666 of 2011-2012

UNDER THE ABLE GUIDANCE OF

DR. KATTAMANCHI HEMACHANDRAN

PROFESSOR

DEPARTMENT OF COMPUTER SCIENCE

ALBERT EINSTEIN SCHOOL OF PHYSICAL SCIENCE

ASSAM UNIVERSITY, SILCHAR

YEAR OF SUBBIMISSION: 2016

CERTIFICATECERTIFICATECERTIFICATECERTIFICATE

This is to certify that the project work entitled “IRIS SEGMENTATION IRIS SEGMENTATION IRIS SEGMENTATION IRIS SEGMENTATION

PROCESSPROCESSPROCESSPROCESS” submitted to Assam University is a bonafide record of the

project carried out by Sahena Begam Barbhuiya in the department of

Computer Science, Assam University, Silchar under my guidance. No part

of the project has been submitted for any other Degree or Diploma .The

work included in this project is original and own work of the candidate.

Place: Silchar Prof. Kattamanchi Hemachandran

Date: (Supervisor)

Department of Computer Science

Assam University, Silchar


SCHOOL OF PHYSICAL SCIENCES

ASSAM UNIVERSITY S IL C HAR

A CENTRAL UNIVERSITY CONSTITUTED UNDER ACT XIII OF 1989

ASSAM, INDIA, PIN - 788011

CERTIFICATECERTIFICATECERTIFICATECERTIFICATE

This is to certify that the project work entitled “IRIS SEGMENTATION IRIS SEGMENTATION IRIS SEGMENTATION IRIS SEGMENTATION

PROCESSPROCESSPROCESSPROCESS” submitted to Assam University is a bonafide record of the

project carried out by Sahena Beagam Barbhuiya in the department of

Computer Science, Assam University,Silchar under the able guidance of Dr.

Kattamanchi Hemachandran, Professor, Department of Computer

Science.No part of the project has been submitted for any other Degree or

Diploma .The work included in this project is original and own work of the

candidate.

Place: Silchar (Dr.Bipul SyamPurkayastha)

Date: HOD


Assam University, Silchar


SCHOOL OF PHYSICAL SCIENCES

ASSAM UNIVERSITY S IL C HAR

A CENTRAL UNIVERSITY CONSTITUTED UNDER ACT XIII OF 1989

ASSAM, INDIA, PIN - 788011


School of Physical Science Assam University, Silchar (A Central University constituted under Act XIII of 1989) Silchar – 788011, Assam, India

DECLARATIONDECLARATIONDECLARATIONDECLARATION

I, Sahena Begam Barbhuiya do hereby declare that the project work entitled “IRIS IRIS IRIS IRIS

SEGMENTATION PROCESSSEGMENTATION PROCESSSEGMENTATION PROCESSSEGMENTATION PROCESS” has been carried out by me under the able guidance of Dr. Dr. Dr. Dr.

Kattamanchi HemachandranKattamanchi HemachandranKattamanchi HemachandranKattamanchi Hemachandran, Professor, Department Of Computer Science, Assam

University, Silchar. This project has not been submitted in any part or full for the award of

any degree in any university or institute.

Place: Silchar (Sahena Begam Barbhuiya)

Date:

ACKNOWLEDGEMENTACKNOWLEDGEMENTACKNOWLEDGEMENTACKNOWLEDGEMENT

At the very outset, I would like to convey my sincere and heartfelt thanks and

gratitude to Dr. Kattamanchi Hemachandran, Professor of Department Of

Computer Science, Assam University, Silchar, for his excellent and able guidance,

valuable suggestions and kind co-operation, which resulted in successful

completion of the project work.

I would like to express my gratitude to Dr.Bipul Syam Purkayastha, Head of

Department of Computer Science, Assam University, Silchar, for his kind co-

operation and help.

I am also thankful to all the respected teachers of the Department of Computer

Science, Assam University,Silchar for their valuable suggestions.

I am pleased to thank the research scholar Sunita Ningthoujam and Arif Iqbal

Mozumder for their great help and co-operation during my project work.

I also wish to express my heartfelt gratitude to the office staff and all other

non teaching staff of the Department of Computer Science, Assam University,

Silchar for their help and support during my project work.

Lastly, I would like to express my deepest regards to my parents, friends and

those who had helped me directly or indirectly in the way to the successful

completion in this aspect.

Place, Silchar Sahena Begam Barbhuiya

Date:

CONTENTSCONTENTSCONTENTSCONTENTS

CHAPTER 1: INTRODUCTION 1-8

1.1 Biometric system 2-3

1.2 Iris Recognition System(IRS) 4

1.3 Stages of Iris Recognition System 4-5

1.4 Advantage of Iris Recognition System 6

1.5 Disadvantage of Iris Recognition System 6

1.6 Application of Iris Recognition System 6

1.7 Motivation 7

1.8 Objectives 8

CHAPTER 2: REVIEW OF LITERATURE 9-14

CHAPTER 3: DESIGN AND IMPEMENTATION 15-22

3.1 Introduction 15

3.2 Pupil detection 15-16

3.3 Iris detection 16-18

3.4 Eyelid detection 18-21

3.5 Noise reduction 21-22

CHAPTER 4: EXPERIMENTAL RESULTS AND DISCUSSION 23-26

CHAPTER 5: USER MANUAL 27-31

CHAPTER 6: CONCLUSION AND FUTURE WORK 32

CHAPTER 7: REFERENCES 33-37

CHAPTER 1

INTRODUCTIONINTRODUCTIONINTRODUCTIONINTRODUCTION

CHAPTER 2

REVIEW OF LITERATUREREVIEW OF LITERATUREREVIEW OF LITERATUREREVIEW OF LITERATURE

CHAPTER 3

DESIGN AND DESIGN AND DESIGN AND DESIGN AND

IMPLEMENTATIONIMPLEMENTATIONIMPLEMENTATIONIMPLEMENTATION

CHAPTER 4

EXPERIMENTAL EXPERIMENTAL EXPERIMENTAL EXPERIMENTAL RESULTSRESULTSRESULTSRESULTS

AND DISCUSSION AND DISCUSSION AND DISCUSSION AND DISCUSSION

CHAPTER 5

USER MANUALUSER MANUALUSER MANUALUSER MANUAL

CHAPTER 6

CONCLUSIONCONCLUSIONCONCLUSIONCONCLUSION AND FUTURE AND FUTURE AND FUTURE AND FUTURE

WORKWORKWORKWORK

CHAPTER 7

REFERENCESREFERENCESREFERENCESREFERENCES

Chapter 1: Introduction

Page: 1

Iris recognition system is one of the most reliable biometric systems for personal

identification due to unique properties of iris and high degree of randomness [1, 2].

Typically, iris recognition systems consist of four modules viz. Iris Segmentation,

Normalization, Feature extraction and Matching [3]. In order to achieve the high

performance iris recognition system, proper segmentation of an iris is required. For

real time application, the computational time of the system should also consider. So,

the segmentation step is the vital steps for overall performance of the system. The

segmentation refers to the isolation of an iris region from an eye image by properly

detecting the iris inner and outer boundaries. Iris region is an annular part between

pupil and sclera as shown in Fig. 1.

Fig. 1 An Iris Image (SI213LOljpg) form CASIA Iris V3 Interval Database

The artifacts such as eyelid, eyelashes, specular reflections etc. make the

segmentation more difficult which in turn decrease the system performance. Daugman

[4] proposed an Integro-Differential operator for proper segmentation of an iris.

Wildes [5], Matveev et al. [6] proposed a segmentation approach based on Hough

transformation. Masek [7] performed the iris segmentation based on Hough

transformation. Circular Hough transform has been applied to detect inner and outer

circle of an iris. In order to improve the speed of the iris segmentation Ma et al. [8],

Zubi et al. [9] roughly determine the iris region in advance before applying Hough

Transform. Shah et al. [10] extracted the iris region from an eye image by using

Geodesic Active contours (GACs). AI-Daoud [11] proposed a method based on

competitive chords to detect pupil-iris and iris-sclera boundaries. Roy et al. [12]


Page: 2

applied parallel game-theoretic decision making procedure to elicit iris boundaries.

Shin et al. [13] applied circular edge detector algorithm to detect the iris boundaries.

Abdullah et al. [14] applied active contour method for complete segmentation of an

iris. Daugman's approach works on local scale and fails to detect circle boundaries

where there is noise such as reflections in the image. Masek's method based on Hough

transformation is computationally expensive to detect the coordinates of the iris [15].

1.1 Biometric System

The term biometrics derived from two Greek words bio means “life” and

metric means “to measure”. It refers to the automatic individual’s identification

based on their physical or behavioral characteristics [16]. The physiological

characteristics of a person such as face pattern, iris pattern, fingerprint, palm print,

hand geometry present unique information to distinguish among them and can be

used in authentication applications [17]. The biometric recognition system

involves two phase viz. Enrollment phase and Identification or Verification phase.

During enrollment phase feature vector is stored in a database after being extracted

from individual object. In identification or verification phase, user provides a

sample vector to the system where it is compare to stored vector and depending on

pre-determined threshold value a decision is made [18]. Fig. 2 demonstrates the

stages of biometric system.

To serve any human physiological or behavioral traits as a biometric characteristic

it should satisfy the following requirements [19, 20]:

• Universality: Every individual should have it.

• Distinctiveness: No two individuals should be the same.

• Permanence. It should be invariable over a given era of time.

• Collectability: The feature must be easy to collect.


Page: 3

Fig 2: Stages in biometric Recognition system [3]

However, for a practical biometric system several other issues [21] such as

performance, acceptability and circumvention of systems are also considered. The

different biometric technologies fingerprint, hand geometry, voice recognition, retinal

recognition, handwritten patterns, iris recognition, and dynamic signature recognition

are used in various applications. Each biometric has its own advantages and

disadvantage and no single biometric system can be considered as an optimal system.

[19].

Due to the advancement of science and technology, application areas of biometrics

are increasing day by day where identification or verification is required for

individuals. The applications of biometrics can be divided into the mainly three

categories viz. Commercial, Government and forensic. Commercial applications

include computer network login, data security, ATM, credit card, distance learning

and many more. Government applications include passport control, national ID and

driver’s license. Criminal investigation, parenthood determination, corpse

identification are belongs to forensic applications of biometric systems. Unlike

traditional methods based on password and PIN number, the use of biometrics

provides more comfort to the user while increasing their security. For example, uses

of biometrics on banking services are much safer and faster compared with existing

methods based on credit and debit cards [22].

Biometric Capture Create Feature Vector

Create Feature Vector

Capture Biometric

STORE

Compare

Match

Enrollment

Verification

Not Match

Threshold


Page: 4

1.2 Iris Recognition System (IRS)

Out of various biometric techniques such as face recognition, fingerprint

recognition, gait, hand and finger geometry, ear, iris recognition have been accepted

as best and most accurate biometric techniques because of the stability, uniqueness,

and non-invasiveness of the iris pattern. The iris region (shown in figure 3), the part

between the pupil and the white sclera provides many minute visible characteristics

such as freckles, coronas, stripes, furrows, crypts which are unique for each

individual. Even two eyes of same person have different characteristics. Furthermore,

the chance of obtaining two people with same characteristics is almost zero that makes

the system efficient and reliable when security is concerned [23].

Fig. 3: Eye Image (CASIA Iris Database)

Typically, the iris recognition system consists of four modules viz. Image

acquisition, segmentation, feature extraction and matching as shown in fig. 4. After

acquiring eye images, iris part is localized by demarcating its inner and outer

boundaries and then the circular iris are transformed to the rectangular with fixed size.

This is done in segmentation and normalization module. Next is the feature extraction

module where the unique iris feature is extracted using appropriate technique from the

segmented iris. Finally, the extracted features are matched with the stored pattern to

validate the identification process [24-25].

1.3 Stages of Iris Recognition System

Iris recognition is a stepwise procedure as shown in fig. 4. The first step is the image

capture. Then images are brought to appropriate forms in order to perform some

preprocessing steps. Then the iris is localized and segmented for further processing.

The texture of the iris is extracted then using appropriate techniques. Finally texture

Iris

Pupil

Eyelid

Sclera


Page: 5

matching is done to validate identification process. The major steps can be

summarized as:

• Image Acquisition: Image is captured under proper illumination, distance

and other factors affecting image quality are taken into consideration. This

step is crucial because image quality plays an important role in iris

Localization.

• Image Segmentation: In this step, the iris region is isolated from the given

image. The iris segmentation is a vital step for overall performance of the

system.

• Feature extraction: The portion of interest i.e. iris patterns are extracted

from the localized iris using techniques like Haar wavelets.

• Matching: The extracted patterns are mapped onto the patterns already

extracted and stored in database. The degree of similarity decides whether

the identification is to be established or not.

Iris Image Acquisition

Iris Segmentation

Normalisation

Feature Extraction

Matching

Compare

Not match

Match

Fig 4: Stages of Iris Recognition System


Page: 6

1.4 Advantage of Iris Recognition System

• The smallest outlier population of all biometrics.

• Iris pattern and structure exhibit long term stability.

• Ideal for handling long databases.

• Comparatively fast matching technique.

• Convenient intuitive user interface.

• Very high accuracy.

• Verification time is generally less than 5 seconds.

1.5 Disadvantage of Iris Recognition System

• Intrusive.

• A lot of memory for the data to be stored.

• Very expansive.

1.6 Application of Iris Recognition System

• One of the most promising applications for iris recognition is that it

increases security for the transportation industry. The current

requirements for security airports could increase the use of biometric

devices in this area.

• Another promising application is for bank ATMs. Someday ATM users

will be identified by their irises rather than their PIN numbers. A

person’s Iris code can be stored either in a database or on a smart card.

• The ability to store the Iris Code on a card or token is important because

it eliminates privacy concerns associated with retaining identities in a

centralized database.

• A biometric technology such as Iris recognition can easily eliminate or

complement the standard log-in password for individual authentication to

a computer.


Page: 7

1.7 Motivation

• Something we hold for security, can be lost, something we know like passwords

or PIN, can be guessed, or forgot. Biometrics provides an alternative to these

methods, or they can be used in combination (multimodal). Fingerprints, which

are widely used, can be forged (gummy fingers). The face changes over a period

of time, even with the best algorithms face recognition(for faces taken one year

apart) has error rates of about 43 to 50 % ,hand geometry is not distinctive enough

to be used in large scale applications, hand-written signatures can be forged.

• Iris recognition systems are the most reliable biometric system since iris patterns

are unique to each individual and do not change with time. A variety of methods

were developed to handle eye data in biometric systems after J. Daugman

developed the first commercial system.

• To obtain a good quality image, users’ cooperation is required like the user must

look straight into camera, still and there should be proper illumination. These

causes inconvenience to the user and is also time consuming.

• Segmentation is a vital step among all the steps involve in iris recognition system.

Proper segmentation of an iris is needed for overall performance of the system.

Various authors proposed different techniques for iris segmentation. All these

existing techniques viz. Daugman nd Masek’s approach are computationally

expensive and time consuming.

• So, in this project work, new segmentation approach has been proposed to

overcome the disadvantages of existing methods.


Page: 8

1.8 Objectives

� To study the various iris recognition systems.

� To study the various iris segmentation techniques.

� Implement an iris segmentation technique for Iris recognition system.

� Compare the proposed segmentation technique with existing techniques in order

to evaluate the efficiency of the proposed technique.

Chapter 2: Review of literature

Page: 9

In the field of iris segmentation techniques, much advancement has been made by

different authors since from 1993, when J.G Daugman proposed an approach for iris

segmentation. In the segmentation stage, this author introduced an Integro-differential

operator to find both the iris inner and outer borders. The methodology used by him is

most popular amongst all iris recognition techniques. In this paper, Daugman assumed

the iris and pupil to be circular and introduced an operator for edge detection. This

operator searches over the image domain (x, y) for the maximum in the blurred

derivative with respect to increasing radius r, of the normalized contour integral of I(x,

y) along a circular arc disk of radius r and center (X0 , Y0)[26].

Wildes [27] proposed an approach for iris segmentation based on intensity

values of the image which is converted into a binary edge map. The edge map is

constructed through the Canny Edge detector. In order to incorporate directional

tuning, the image intensity derivatives are weighted to favour ranges of orientation.

Then the well-known Circular Hough Transform is used to obtain the boundaries. The

accuracy of this methodology is dependent on the edge detection algorithm.

Over the past decade, Daugman [28] has constantly modified and improved the

recognition algorithms. In [28], Daugman presented alternative segmentation methods

based on active contours to transform an off-angle iris image into a more frontal view.

Here, to deal with non-circular iris inner and outer boundaries discrete Fourier

transform is applied to detect the iris inner and outer boundaries.

Boles et al. [29] developed an iris recognition system using 1-D dyadic wavelet

transform with various resolution levels on an iris image to characterize the texture of

the iris and then used zero-crossing for feature representation. It made the use of two

dissimilarity functions to compare the new pattern with the reference patterns. Boles’

approaches have the advantage of processing 1-D iris signals rather than a 2-D image.

Here, ‘‘1D dyadic’’ means a pair of 1 dimensional wavelet filters, such as low pass

and high pass filters.

Li Ma et al. [30] defined new spatial filters to extract iris features, which were

Gaussians modulated by circularly symmetric sinusoidal function. Experiments show


Page: 10

that the method was suitable for iris feature extraction. Daouk et al. [31] used the

fusion of the canny edge detection scheme and circular Hough transform for iris

segmentation. Based on the Gabor filters and the characteristics of the iris pattern,

Chen et al. [32] introduced Gradient Direction Coding (GDC) with grey code method

and Delta Modulation coding (DMC) method to extract and encode the iris features.

GDC method is a 2-D method and encodes the gradient direction of each small 2-D

iris image block in Wavelet Transform domain. DMC method is delta modulation

concept used to efficiently encode the feature information. This is a 1-D method in

which the 2-D feature information is converted into 1-D feature signals before

encoding.

Barzegar et al. [33] proposed an iris segmentation method based on point wise

level set approach. Yahya et al. [34] proposed efficient technique for iris localization.

First, Direct least square fitting of the ellipse was used to detect the iris inner

boundaries and then Integro-differential operator was applied to detect outer

boundaries which increase the speed and accuracy in iris segmentation compare to

other approaches. Abra et al. [35] proposed an algorithm based on optical composite

correlation filter. The algorithm eliminates redundancy by using new design of

composite filter called Indexed Composite Filter (ICF). The values of the inner and

the outer boundary are determined through two ICF.

Murty et al. [36] presented a new and modified algorithm for iris recognition based

on canny edge detection scheme and Hough transform to segment the iris part. The

segmented iris part was normalized by using the rubber sheet model and then

Principal component analysis (PCA) was used for pattern matching. Yahya et al. [37],

introduced a Chan-Vese active contour method to extract the iris. The reflections were

identified by using impainting technique in loaded image. The adaptive boosting

(AdaBoost)-Cascade Detector was adopted to detect iris region. Finally Chan-Vese

active contour method was applied to find the iris boundaries. It performed better with

an error rejection rate (ERR) of 5.5068 compare to Daugman of 16.8635 and Wildes

of 33.8226 when the UBIRIS iris database is considered.

Most commercially available iris recognition systems are based on the pioneered

algorithms of Daugman [28] and Wildes [27]. However, they perform well in ideal


Page: 11

conditions but may fail for non-ideal data. The non-ideal eye images may contain

multiple issues such as specular reflections, low contrast, blurring, focus, non-uniform

illumination, glasses and contact lens, off-axis and the off-angle eyes; occlusions such

as eyelashes, eyelids and hair [27].

Jan et al. [38] proposed a robust iris localization algorithm for non-ideal eye image

based on the Hough transform, gray level statistics, adaptive thresholding and a

geometrical transform. The algorithm involves two phases. In the first phase, iris

circle in a sub-image centered at the pupil circle was localized after localizing the

pupil region. However, on failure of first phase the coarse iris region was localized in

the second phase. Finally, the iris circular boundaries were regularized by using radial

gradients and the active contours. The proposed technique was tolerant to off-axis eye

images, specular reflections, non-uniform illumination, glasses, and contact lens, hair,

eyelashes, and eyelids occlusions.

Li et al. [39] introduced a robust iris segmentation algorithm based on the

Random Sample Consensus (RANSAC). The algorithm localized the iris boundaries

more accurately than the methods based on the Hough transform. Li et al. [40]

presented a weighted co-occurrence phase histogram (WCPH) for representing the

local characteristics of texture pattern which accounts for inconsistencies brought by

the disturbing factors such as noise, illumination changes. Raffei et al. [41] proposed

an algorithm based on a multi-scale sparse representation of local radon transform

(msLRT) to extract iris features when eye images were captured in a non-cooperative

environment and under visible wavelength illumination. The method was able to

reduced the computational complexity and to generate a compact iris feature vector.

Jan et al. [42], introduced reliable iris localization techniques using Hough

transform, histogram bisection and eccentricity for non-ideal iris image. It includes

localizing a coarse iris location in the eye image using the Hough transform and image

statistics; localizing the pupillary boundary using a bi-valued adaptive threshold and

the two-dimensional (2D) shape properties; localizing the limbic boundary by reusing

the Hough accumulator and image statistics; and finally, regularizing these boundaries

using a technique based on the Fourier series and radial gradients. The experimental

result shows that the proposed method has tolerance to non-ideal issues, such as the


Page: 12

off- axis eye images, specular reflections; hair, glasses, cosmetic lenses, eyelids, and

eyelashes occlusions. Sun et al.[43] approach was based on scale invariant feature

transform (SIFT) and bag-of-features. After detecting the iris inner boundary by

region based active contour, SIFT method is applied to detect the key points in the iris

image. The points located in pupil region were removed. The histogram representation

for each iris image was generated from the constructed feature vocabulary. The

histogram distance was adopted for the matching test.

Chowhan et al. [44] described the Modified Fuzzy Hyperline Segment Neural

Network (MFHSNN) based iris recognition. The Gabor filters technique was used for

Iris feature extraction after segmentation and normalization using Integro-differential

operator and Cartesian to Polar Coordinate transform respectively. The FHLSNN with

its learning algorithm was used for classification of iris patterns.

Bindra et al. [45] introduced an iris recognition system by dividing the iris

image into three and two to reduce the computational complexity which was different

from a traditional system where complete iris image is extracted. The Sobel operator

and wavelet transformation were used for feature extraction. The combination of the

Euclidean distance with Particle Swarm Optimization (PSO) was presented for

classification. The algorithm was tested on an IITK iris database. Logannathan et al.

[46] presented an approach based on wavelet probabilistic neural network (WPNN)

model. WPNN model combines the wavelet neural networks and probabilistic neural

networks (PNN) and able to improve the recognition accuracy and system

performance. The PSO technique was used to train the WPNN. PSO is an

evolutionary computation technique developed and can search automatically the

optimum solution in the vector space.

Poornima et al. [47] presented a new way of iris segmentation based on neural

network. The neural network was trained with best iris localization method among

Daugman’s algorithm, Hough Transform, Canny edge detection algorithm and

Integro-differential operator. The best method was selected based on output of each

algorithm. The Integro-differential operator was found better than other algorithms

and used to train the network. Other neural network techniques such as Intersecting


Page: 13

Cortical Model (ICM) Neural Network to generate iris code, Notation Spreading

Neural Network (R-SAN net) are also proposed for iris recognition. The ICM neural

network is a simplified model of pulse-coupled neural network (PCNN) has excellent

performance for image segmentation whereas the R-SAN network is suitable for

recognizing the orientation of the object regardless of its shape [48]. The survey on

iris recognition systems summarized in table 1.

Table 1: Iris Recognition techniques

References Description

Segmentation and Normalization Feature Extraction & Matching

Daugman

1994 [26]

Integro-Differential operator and Rubber

Sheet Model

2D Gabor wavelet and Humming Distance

Boles et al.

1998 [29]

Edge detection schemes Zero crossing based on Dyatic wavelet

transform and dissimilarity function

Daouk et al.

2002 [31]

Canny edge detection scheme, circular

Hough transform and Bilinear

Transformation

Haar wavelet transform for feature

extraction and Humming distance for

pattern matching

Chen et al.

2005 [32]

Edge detection algorithm Wavelet transform & Humming distance

Daugman

2007 [28]

Active Contour methods Gabor wavelets & Humming distance

Barzegar et

al. 2008

[33]

Point wise level set approach Not described

Yahya et al.

2008 [34]

Direct least square fitting of ellipse and

Integro-differential operator

Not described

Abra et al.

2009 [35]

Indexed composite correlation filters indexed composite phase only filter

(ICPOF)

Murty et al.

2009 [36]

Canny edge detector, Hough Transform

& Rubber Sheet Model

Gabor filter and Principle Component

Analysis (PCA)

Yahya et al. Chan-Vese active contour method 1-D log polar Gabor transform &


Page: 14

2010 [37] Humming distance

Jan et al.

2012 [38]

Hough Transform, bi-valued adaptive

threshold, & the Fourier series and radial

gradients.

Not described

Li et al.,

2012 [39]

Random Sample Consensus (RANSAC) Gabor filter

Li et al.,

2012 [40]

Canny edge detection Algorithm,

circular Hough transform & Rubber

sheet model

weighted co-occurrence phase histogram

(WCPH)& Bhattacharyya distance

Jan et al.

2013 [38]

Hough transform, gray level statistics,

adaptive thresholding, and a geometrical

transform

Not described

Sun et al.,

2013 [43]

Scale Invariant Feature Transfor &

Active contour model

Histogram Distance

Chowhan et

al. , 2011

[44]

Integro-differential operator & Rubber

sheet model

2D spatial Gabor wavelet filters &

Modified Fuzzy Hyperline Segment

Neural Network

Bindra et

al., 2012

[45]

2-D wavelet filtering & Rubber sheet

model

Sobel Operator and 1-D wavelet transform

& Euclidean Distance

Logannatha

n et al.,

2012 [46]

Hough Transform & Sobel Transform Wavelet Probabilistic Neural Network

(WPNN) with Particle swam optimization

as training algorithm.

Poornima et

al., 2010

[47]

Artificial neural network with Integro-

differential operator

Not described

Chapter 3: Design and Implementation

Page: 15

3.1 Introduction

The personal identification based on Iris biometric is one of the most suitable

and reliable methods with respect to performance and accuracy. However, the

reliability and accuracy of the method depends on the proper segmentation of an iris

from an eye image. Typically, iris recognition systems consist of four modules viz.

Iris Segmentation, Normalization, Feature extraction and Matching [49]. In order to

achieve the high performance iris recognition system, proper segmentation of an iris is

required. For real time application, the computational time of the system should also

consider. So, the segmentation step is the vital steps for overall performance of the

system. The segmentation refers to the isolation of an iris region from an eye image by

properly detecting the iris inner and outer boundaries. In this project work, new Iris

segmentation Approach has been proposed. The Fig. 5 shows the proposed

segmentation approach for iris segmentation.

3.2 Pupil Detection

The first step of the proposed segmentation approach is to localize the pupil. With

the help of pupil location, image is cropped into a rectangle which holds the whole eye

image. The Pupil localization is done by thresholding. The following algorithm 1

describes the one process of pupil localization and boundary detection from an eye

image.

Algorithm 1:

i. Create binary image with threshold value ≤ 30

Fig. 5: Proposed Segmentation Approach

Input Image

Iris Detection

Eyelid Detection

Noise Detection

Segmented Iris

Pupil Localization

Crop Image

Pupil Detection


Page: 16

ii. Perform morphological operation ‘detect’ and ‘erode’ in following by

flood fill.

iii. Apply median filter in order to remove smaller region.

iv. Determine the centroid of the detected pupil region (Cx , Cy)

v. The horizontal radius.

The result of the pupil detection algorithm on the input image is shown in

fig.6

(a) (b) (c) (d) (e)

(f)

3.3 Iris Detection

In this project work, the Circular Hough Transform (CHT) described in algorithm

2 has been applied on the cropped image to detect the iris outer boundary.

Circular Hough Transform is used to detect the presence of circular shapes in any

image. For example detection of number of circular discs in an image. Another well

Fig. 6 (a) Original Image (b) Binary image (c) Image after flood fill (d) Image after morphological operations (e) Image after median filtering (f) Detected pupil region


Page: 17

known application of Circular Hough Transform is the detection of number of

coconuts in an image [50]. Circular Hough Transform uses the parameterized equation

of circle for this purpose.

The equation of circle can be written as

(� − �)� + (� − )� = ��

Where x, y are the points on the circumference of the circle, a, b is the centre of circle

and r is the radius of the circle.

The equation of circle can be written as

� = � + � ∗ cos(�)

� = + � ∗ sin(�)

The circular Hough Transform uses these equations to compute the CHT of a

circle and detect the presence of circular objects in an image. The CHT algorithm used

in this project work is defined as:

Algorithm 2:

i. Read an image file.

ii. Find edges in the image.

iii. Define a radius range to be used.

iv. For each edge point, draw a circle with that edge point as the centre and a

radius r and increment the number of votes by 1 for all the coordinates that

coincide with the circumference of the circle drawn, in the accumulator

space and find circles for an edge point for all the radius in the range.

v. Find maximum number of votes in the accumulator space.

vi. Plot circle with parameters (r, a, b) corresponding to the maximum votes in

the accumulator space.

vii. The circle obtained is the desired circle with (r, a, b) as the radius and

centre of circle respectively.


Page: 18

The result of iris detection step is on different images of the considered dataset is

shown in figure 7.

3.4 Eyelid Detection

In most of the cases, top and bottom eyelid overlaps the iris region. For this proper

detection and isolation of eyelids is necessary for accurate iris segmentation. In

this project work, canny edge detection method and circle geometry has been

applied in order to isolate the eyelids.

The top and bottom eyelids are isolated by using Algorithm 3 and Algorithm 4

respectively

Algorithm 3: Top Eyelid Detection

i. Select three small rectangles from the top side of the given image, two on

left and right of detected iris and one on top of the detected pupil, as shown

in fig 8.

Fig. 7: Detected Iris Boundaries

Fig. 8: Image with Selected rectangles


Page: 19

ii. Apply adaptive histogram equalization and Median Filtering to enhance

the selected portion.

iii. Detect the horizontal line by using canny edge detection technique on each

of the rectangle and calculate middle point of the detected line as shown in

figure 9.

The detected points will lie on the top eyelid edge.

iv. The curves passing these detected points are drawn using circle geometry.

Circle geometry states that there is one and only are circle passing through

the any three non-collinear points.

The result of Algorithm 3 is shown in fig. 10.

Fig. 9: Image with detected points

Fig. 10: Image with detected top eyelid


Page: 20

Algorithm 4: Bottom Eyelid Detection

i. Select three small rectangles from the bottom side of the given image, two

on left and right of detected iris and one on bottom of the detected pupil, as

shown in fig 11.

ii. Apply adaptive histogram equalization and Median Filtering to enhance

the selected portion.

iii. Detect the horizontal line by using canny edge detection technique on each

of the rectangle and calculate middle point of the detected line as shown in

figure 12.

The detected points will lie on the bottom eyelid edge.

Fig. 11: Image with selected rectangles

Fig. 12: Image with detected points


Page: 21

iv. The curves passing these detected points are drawn using circle geometry.

The result of Algorithm 3 is shown in fig. 13.

3.5 Noise Reduction

Eye lashes and reflections are removed from the segmentation iris by linear

thresholding method. The Pseudo-code of thresholding method is as follows:

[row, col]=size(img); for i=1:row for j=1:col if ((j-xt)^2+(i-yt)^2)>rt^2 img(i,j)=255; end if ((j-xb)^2+(i-yb)^2)>rb^2 img(i,j)=255; end if ((j-xp)^2+(i-yp)^2)<rp^2 img(i,j)=255; end if ((j-xi)^2+(i-yi)^2)>ri^2 img(i,j)=255; end end end for i=1:row for j=1:col if ((j-xp)^2+(i-yp)^2)>rp^2 if ((j-xi)^2+(i-yi)^2)<ri^2 if img(i,j)>240; img(i,j)=255; img(i,j)=255; end if img(i,j)<=30; img1(i,j)=255; img(i,j)=255; end end

Fig. 13: Image with detected bottom eyelid


Page: 22

end end end

where (xt, yt), (xb, yb), (xp, yp) and (xi, yi) are the centre coordinates of top eyelid,

bottom eyelid, pupil and iris respectively. And rt, rb, rp and ri are the radius of the top

eyelid, bottom eyelid, pupil and iris respectively. The threshold value 30 and 240

selected by trial and error method. The result of whole iris segmentation approach on

various images is shown in fig 15.

a b c d

Fig 14: (a) Original Image (b) Localized Iris (c) Detected top and Bottom eyelid (d) Segmented Iris

Chapter 4: Experimental result and Discussion

Page: 23

The experiments were implemented in MATLAB 7.12.0 and executed on Intel Core i3

2.4 GHz with 3 GB RAM. To evaluate the performance of the approach CASIA Iris

lamp database is used [51]. This data base consists of 16440 images from 411 persons.

Each image is an 8 bit gray level value of resolution 480*640. Total 250 images of left

eye of 50 persons are taken randomly for the experimentation. The results of the

proposed iris segmentation approach on different eye images are shown in fig. 16.

The performance of the proposed approach has been evaluated in terms of running

time of the system and segmentation accuracy. To validate the results obtained by the

proposed approach, results are compare with existing method viz. Daugman’s method

[26] and Masek’s method [52].

c d b a

Fig 15: (a) Original Image (b) Localized Iris (c) Detected top and Bottom eyelid (d) Segmented Iris


Page: 24

The average running time of the system with proposed approach and with existing

methods are shown in table 1. From this table, it is observed that the proposed

segmentation method is efficient compare to other methods viz. Daugman and Masek

with respect to the running time for iris segmentation.

To calculate the segmentation accuracy, the radius and centroid of the iris and

pupil is calculated manually with the help of MATLAB Image tool and circle

Table 1: Running time for iris segmentation

Method Technique

Running time of iris

Segmentation

Min. time

(sec.)

Max.

time

(sec.)

Avg. time

for 200

images

(sec.)

Masek [52]

Modified Canny and circular Hough

Transform is used to detect inner and outer

circle of an iris.

Isolation of top and bottom eyelids.

Linear thresholding techniques to isolate

reflection and eyelashes.

7.28 37.94 19.47

Daugman

[26]

Integro-differential operation is used to

detect iris inner and outer boundary. 10.45 61.10 27.70

Proposed

Approach

Pupil localization by simple morphological

operations with the help of thresholding and

median filtering technique.

Iris localization by Circular Hough

Transform.

Eye lashes and reflections are removed by

linear thresholding.

Top and bottom Eyelid is isolated using

canny and Circle Geometry.

1.9 4.35 3.11


Page: 25

geometry. The fig. 2 shows the difference between manually computed and system

computed iris boundaries. The segmentation accuracy is calculated as:

�� = 1 − ��(�� − ��)��

+ ��(�� − ��)��

� × 100

Where �� and �� represents the radius of iris outer boundary and pupil computed

manually respectively. The �� and �� represents the radius of iris outer boundary and

pupil computed by the system respectively. Table 2 shows the segmentation accuracy

of the proposed method on different from the iris image database.

Fig 16: The difference between manually computed and iris boundaries computed by

proposed method. Red circles represent manually computed Iris Boundaries and blue circles

represent iris boundaries computed by proposed approach

Image Segmentation Accuracy

‘S2001L01.JPEG’ 98.63%

‘S2002L01.JPEG’ 97.43%

‘S2003L06.JPEG’ 96.95%

‘S2003L12.JPEG’ 99.33%

‘S2003L13.JPEG’ 98.91%

Table 2: Segmentation Accuracy


Page: 26

The average accuracy of the proposed method is 98.20 %. From these results, it

is observed that the proposed segmentation approach achieved high degree of

segmentation accuracy in a reasonable amount of time and it is suitable for

accommodate in a real life iris recognition system.

Chapter 5: User Manual

Page: 27

• The following screen is the starting screen which has five button options: load

image, localize iris, localize eyelids, remove noise and clear.


Page: 28

• To load an image, click on the “Load image” button and a dialogue box will

display which contains a folder of various eye images. Select and open the file

folder for displaying the eye images as shown below :

• Select the image file which you want to segment and click on “Open” button.


Page: 29

• Selected eye image file will be displayed in the “Original Image” field.

• To localize the pupil and iris outer boundary, click on “Localize Iris” button

and image will be displayed in the “Localized Iris” field as shown below:


Page: 30

• To detect the top and bottom eyelids, click on “Localize Eyelids” button. The

detected top and bottom eyelids will be displayed in the “Detected Eyelids”

field.

• To remove the noises from the eye image click on “Remove Noises” button

and the display output will be the “Segmented Iris” of the selected eye image

along with its “Segmentation Time” as shown below:


Page: 31

• Click the “Clear” button to clear all the fields.

Chapter 6: Conclusion and Future work

Page: 32

In this project work, new Iris segmentation approach has been proposed. Simple

morphological operations and two dimensional median filtering techniques are used to

detect the pupil. The Iris outer boundary is detected by using circular Houghman

Transformation (CHT). The canny edge detection method with circle geometry is

applied to isolate upper and lower eyelids. The noises such as eyelashes, reflections

are removed through the linear thresholding. From the experimental results, it is

observed that the proposed method is more efficient compare to existing methods viz.

Daugman and Masek for the considered dataset. It is also observed that proposed

method takes reasonable amount of time to perform iris segmentation.

The future work would be to test the influence on accuracy of the proposed

method over a large dataset and also to develop the iris recognition system with this

segmentation approach to validate the proposed method on recognition accuracy.

Chapter 7: References

Page: 33

[1] S. A. Sahmoud, et al., “Efficient iris segmentation method in unconstrained

environments”, Pattern Recognition, Elsevier, pp. 3174-3185, 2013.

[2] T. Lefevre, et. al., “Effective Elliptic Fitting for Iris Normalization”, Computer Vision

and Image Understanding, Elsevier, 2013.

[3] S. Sun., et al., “Non-cooperative Bovine Iris Recognition via SIFT”,

Neurocomputing, Elsevier, pp. 310-317, 2013.

[4] J. G. Daugman, “High Confidence Visual Recognition of Persons by a Test of

Statistical Independence”, IEEE Trans, Pattern Anal. Mach Intell, vol. 15, pp. 1148-

1161, 1993.

[5] R. P. Wildes, “Iris Recognition: An Emerging Biometric Technology”, Proc. IEEE,

Vol. 85(9), pp. 1348-1363, 1997.

[6] I. Matveev, “Iris segmentation system based on approximate feature detection with

subsequent refinements”, Computer Society, IEEE, pp. 1704-1709, 2014.

[7] Libor Masek, Peter Kovesi. MATLAB Source Code for a Biometric Identification

System Based on Iris Patterns. The School of Computer Science and Software

Engineering, The University of Western Australia. 2003.

[8] L. Ma, Y. Wang, D. Zhang, “Efficient iris recognition by characterizing key local

variations”, IEEE Trans. Image Process. Vol. 13(6), pp. 739-750, 2004.

[9] R. T. A. Zubi, D. I. A. Nadi, “Automated personal identification system based on

human iris analysis”. Pattern Anal Applic, pp. 147-164, 2007.

[10] E. A. Daoud, “A New Iris Localization method based on the competitive chords”,

SlVip, 6, pp. 547-555, 2012.

[11] S. A. Sahmoud, I. S. Abuhaiba, “Efficient iris segmentation method in unconstrained

environments”, Pattern Recognition, Elsevier, 2013.

[12] K. Y. Shin, et al., “New Iris Recognition Method for Noisy Iris Images”, Pattern

Recognition Letters, 33, pp. 991-999, 2012.


Page: 34

[13] X. He, P. Shi, “A new segmentation approach for iris recognition based on hand-held

capture device”, Pattern Recognition, Vol. 40, Elsevier, pp. 1326-1333, 2007.

[14] M. P. Stephen, et al., “Adaptive histogram equalization and its variations”, Computer

Visions, Graphics and Image Processing, Vol. 39, pp. 355-368, 1987.

[15] A. Walid, K Lotfi, M. Nouri, “A Fast and Accurate Eyelids and Eyelashes Detection

Approach for Iris Segmentation”, Journal of Multimedia Processing and

Technologies, Vol. 3 No. 4,pp. 166-173, 2012.

[16] Nabti M. et al.: An effective and fast iris recognition system based on a combined

multi-scale feature extraction technique. Pattern Recognition, Elsevier, 41, 2008, pp.

868-879.

[17] Alice I: Biometric Recognition: Security and Privacy concern, IEEE Security and

Privacy, 2003.

[18] Ramkumar R. P.: Novel Iris Recognition Algorithm. ICCCNT, IEEE, 2012, pp. 1-6.

[19] Jain A. K., et al.: An Introduction to Biometric Recognition. Trans. IEEE, Circuits

and Systems for Video Technology, 14, 2004, pp. 4-20.

[20] Huang J.: A New Iris Segmentation Method for Recognition. In: Proc. of the 17th

International Conference on Pattern Recognition, IEEE, 2004.

[21] Jain A. K.: An Introduction to Biometric Recognition. Trans. on Circuits and Systems

for Video Technology, IEEE, 14, 2004, pp. 4-20.

[22] Abiyev R. H., Altunkaya K.: Personal Iris Recognition Using Neural Network.

International Journal of Security and its Applications, 2, 2008, pp. 41-50.

[23] Jhamb M. et al.: IRIS Based Human Recognition System. Int. Journal of Biometrics

and Bioinformatics (IJBB), 5, 2011, pp. 1-13.

[24] Daugman J. G.: High Confidence Visual Recognition of person by a test of Statistical

Independence. Trans. on Pattern Analysis and Machine Intelligence, IEEE, 15, 1993,

pp. 1148-1161.


Page: 35

[25] Devi, Ningthoujam Sunita, and K. Hemachandran. "Automatic Face Recognition

System using Pattern Recognition Techniques: A Survey."

[26] Daugman J. G.: Biometric Personal Identification System Based on Iris Analysis,

United States Patent, no. 5291560, 1994.

[27] Wildes R. P.: Iris Recognition: An Emerging Biometric Technology. In: Proc. of the

IEEE, 85, 1997, pp. 1348-1363.

[28] Daugman J. G.: High Confidence Visual Recognition of person by a test of Statistical

Independence. Trans. on Pattern Analysis and Machine Intelligence, IEEE, 15, 1993,

pp. 1148-1161.

[29] Boles W., Boashash B.: A Human Identification Technique Using Images of the Iris

and Wavelet Transform. IEEE Trans. Signal Processing, 46, 1998 pp.1185-1188.

[30] Ma L. et al.: Personal Identification Based on Iris Texture Analysis. Trans. on Pattern

Analysis And Machine Intelligence, IEEE, 25, 2003, pp.1519-1533.

[31] Daouk C. H. et al.: Iris Recognition, In: Proc. of the 2nd IEEE Int. Symposium on

Signal Processing and Information Technology, 2002, pp. 558-562.

[32] Chen W. S. et al.: Personal Identification Technique based on Human Iris

Recognition with Wavelet Transform. Int. Conf. on Acoustics, Speech and Signal

Processing, IEEE, 2, 2005, pp. ii - 949.

[33] Barzegar N. et al.: A New Approach for Iris Localization in Iris Recognition Systems.

, Int. conf. on Computer Systems and Applications, IEEE, 2008, pp. 516-523.

[34] Yahya A. E. et al.: A New Technique for Iris Localization in Iris Recognition

Systems. Information Technology Journals, 7, 2008, pp. 924-929

[35] Abra O. E. K. et al.: Optical Iris Localization Approach. In: proc. of the IEEE Int.

Conf. on Computer Systems and Applications, 2009, pp. 563-566.

[36] Murty P. S. R. C. et al.: Iris Recognition System using Principal Component of

Texture Characteristics. International Journal of Computing Science and

Communication Technologies, 2, 2009, pp. 343-348.


Page: 36

[37] Yahya A. E. et al.: Accurate Iris Segmentation Method for Non-Cooperative Iris

Recognition System. Journal of Computer Science, 6, 2010, pp. 492-497.

[38] Jan F. et al.: Iris localization in frontal eye images for less constrained iris recognition

systems, Digital Signal Processing, Elsevier, 22, 2012, pp. 971-986.

[39] Li. P. et al.: Iris Reognition in non-ideal imaging conditions, Pattern Recognition

Letters, Elsevier, 33, 2012, pp. 1012-1018.

[40] Li. P. et al.: Weighted Co-occurrence Phase Histogram for Iris Recognition, Pattern

Recognition Letters, Elsevier, 33, 2012, pp. 1000-1005.

[41] Raffei A. F. et al.: Feature Extraction for Different Distances of Visible Reflection

Iris Using Multiscale Sparse Representation of Local Randon transform, Pattern

Recognition, Elsevier, 46, 2013, pp. 2622-2633.

[42] Jan F. et al.: Reliable Iris Localization Using Hough Transform, Histogram-bisection,

and Eccentricity. Signal Processing, Elsevier, 93, 2013, pp. 230-241

[43] Sun S. et al.: Non-cooperative bovine iris recognition via SIFT. Neurocomputing,

Elsevier, 120, 2013, pp. 310-317.

[44] Chowhan S. S. et al.: Iris Recognition Using Modified Fuzzy Hyperline Segment

Neural Network. Journal of Computing, 3, 2011, pp. 72-77.

[45] Bindra G. S. et al.: Feature Based Iris Recognition System functioning on Extraction

of 2D Features. International Conference on System Engineering and Technology,

IEEE, 2012, 1, pp. 17-19.

[46] Logannathan B. et al.: Iris Authentication Using PSO, International Journal of

Computer& Organization Trends, 2, 2012, pp. 10-15.

[47] Poornima S.: Comparison and a Neural Network Approach for Iris Localization.

Procedia Computer Science, Elsevier, 2, 2010, pp. 127–132.

[48] Proenca, H.: Iris segmentation methodology for non cooperative recognition. In: Proc.

of Vis. Image Signal Process, IEE, 153, 2006, pp. 199-205.


Page: 37

[49] Kumar V. et al.: Importance of Statistical Measures in Digital Image Processing.

International Journal of Emerging Technology and Advanced Engineering, 2, 2012,

pp. 56-62.

[50] Daugman J.: New Methods in Iris Recognition. Trans. on Systems, Man, and

Cybernatics - Part B: Cybernatics, IEEE, 37, 2007, pp. 1167-1175.

[51] Huang Y. et al.: An Efficient Iris Recognition System. In: Proc. of the First Int. Conf.

on Machine Learning and Cybernetics, IEEE, 1, 2002, pp. 450-454.

[52] Zhou Z. et al.: A New Iris Recognition Method Based on Gabor Wavelet Neural

Network. Int. Conf. on Intelligent Information Hiding and Multimedia Signal

Processing, Computer Society, IEEE, 2008, pp. 1101-1104.

IRIS SEGMENTATION PROCESS - csnow.in SEGMENTATION PROCESS ... Kattamanchi Hemachandran, ... I would...

Documents

Transcript of IRIS SEGMENTATION PROCESS - csnow.in SEGMENTATION PROCESS ... Kattamanchi Hemachandran, ... I would...