International Conference on Electrical, Electronics, and Optimization Techniques (ICEEOT) - 2016
Evolution of Performance Analysis of Iris
Recognition System By using Hybrid Methods of
feature Extraction and Matching by Hybrid Classifier
for Iris Recognition System
Aparna G. Gale Dr. Suresh S. Salankar
Electronics & Telecommunication Dept. Professor, Electronics & Tecommunication Engg. Dept.
Om College Of Engg. G.H.Raisoni College Of Engg.
Wardha, India Nagpur, India
aparna.gale@gmail.com suresh.salankar@gmail.com
Abstract— In today’s world the higher stable and distinct to be much safer, faster and easier than the existing methods
biometric characteristics to identify and / or to verify any person are based on credit and debit cards. Proposed forms of payments
the human iris. Iris recognition system consists image acquisition, such as pay and touch scheme based on fingerprint or smart
localization, normalization, features extraction and matching. Iris cards with stored iris information on them are the examples
images are taken from CASIA iris VI database for study. In this
of such applications. Biometric systems are widely used for
paper we make a comparative study of performance of image
transform using Haar transform, PCA, Block sum algorithm and
authentication, identification and verification of any
hybrid algorithm for iris verification to extract features on specific individual. In terms of accuracy, face, fingerprint and iris
portion of the iris for improving the performance of an iris based system are considered to be most effective. Since
recognition system. The hybrid methods are evaluated by combining fingerprint of an individual changes over time and face
Haar transform and block sum algorithm. The classifiers used in recognition systems requires large database area and high
this study are hybrid classifier i.e. ANN and FAR/ FRR and the matching time. They are considered infeasible for high
experimental results show that this technique produces good accuracy, large size recognition application. Iris texture of an
performance on CASIA VI iris database. individual remains stable through life and can be encoded in
small memory. These features make iris based recognition
most accurate and reliable biometric identification available.
Keywords - Biometric; Iris Recognition; Haar transform; ANN;
Block sum algorithm; PCA.
II.BASIC STEPS OF IRIS RECOGNITION SYSTEM
I. INTRODUCTION
The iris is a thin circular diaphragm which lies between the
Biometric deals the method of recognizing individuals cornea and the lens of the human eye. The front view of the
based on their physiological or behavioral characteristics. iris shown in fig. 1.
The physiological characteristics are based on the biological
individuality of users like fingerprints, face, hand geometry,
vein patterns, retina and iris. The behavioral characteristics
consider voice, handwritten signature.
The development in science and technology has made it
possible to use biometrics in application where it is required
to establish or to conform the identity of individuals.
Applications such as control, database access and financial
services are some passenger control in airports, access control
in restricted areas, border of the examples where the
biometric technology has been applied for more reliable
identification and verification.
In the field of financial services, biometric technology
Fig.1 Front view of human eye
has shown a great potential in offering more comport to
customers while increasing their security. As an example The Basic steps of Iris recognition system are as shown in
banking services and payments based on biometrics are going fig. a.
978-1-4673-9939-5/16/$31.00 ©2016 IEEE
3259
Image Acquisition
Image Localization
Image Fig.3. Original Image Fig. Expected Image
Preprocessing
Image Normalization The pupil boundary algorithm was applied over the image
mage to retrieve the pupil boundary by setting threshold
value lowest intensity 0 and for limbic boundary highest
intensity value 255. The Canny edge detection algorithm was
Feature Extraction Feature Extraction By Feature applied to retrieve the edges and the inbuilt Gaussian filtering
By Haar Transform PCA Extraction By
Block Sum
helps to retrieve the smooth and sharpened image edges.
2.2.2 Iris Normalization:
Irises from different people may be captured in different
Matching By ANN size and even for irises from the same eye the size may
change due to illumination variation and other factors. Such
Match
Decision By
elastic deformation in iris texture will affects the result of iris
FAR/FRR matching. For the purpose of achieving more accurate
No recognition results, it is necessary compensate for the iris
Match
deformation.
FIG.2. IRIS RECOGNITION SYSTEM 2.3. Feature Extraction
2.1 Image Acquisition Feature extraction identifies the most prominent features
for classification Iris provides abundant texture information.
Iris image acquisition is the first step in iris recognition.
A feature vector is formed which consists of the ordered
The small size of iris combined with the possibility of
sequence of feature extracted from the various representation
varying iris colors means a special camera must be used
of the iris images. Some of the features are X-Y coordinates,
especially for people with darker colored irises. A good and
radius, shape & size of the pupil and ratio between average
clear image eliminates the process of noise removal and also
intensity of two pupils. Here we have taken three algorithms
helps in avoiding errors in calculation. This paper uses the
for feature extraction.
image provided by CASIA database. These images were
taken solely for the purpose of iris recognition software 2.3.1 Feature Extraction with Principal Component
research and implementation. Analysis.
2.2. Image Preprocessing:
The aim of feature extraction is to find a transformation
The preprocessing of eye image is essential for getting the from an n-dimensional observation space to a smaller m
required and accurate input for further processing. Image dimensional feature space. Main reason for performing
capturing is 1st step and quality of input image helps to store feature extraction is to reduce the computational complexity
the biometric distinctive feature extraction easier and faster. for iris recognition. Most existing iris recognition methods
Images can be obtained from database. The preprocessing of are based on the local properties such as phase, shape, and so
the iris image is the combination of image localization & on. However, iris image recognition based on local properties
normalization. is difficult to implement. Principal component analysis can
produce spatially global features. The original data are thus
2.2.1 Image Localization: projected onto a much smaller space, resulting in data
Iris localization is the process of finding the iris lower reduction.PCA was invented in 1901 by Karl Pearson.
and upper boundary value. The input image is shown in fig. Principal component analysis (PCA) is a classic technique
The expected image of the iris boundaries are shown in fig. used for compressing higher dimensional data sets to lower
c. dimensional ones for data analysis, visualization, feature
extraction, or data compression. PCA involves the calculation
of the eign value decomposition of a data covariance matrix
or singular value decomposition of a data matrix, usually
after mean entering the data for each attribute.
3260
2.3.2 Feature extraction using Block sum: 2.3. For every Database image „iಲand a Query image „qthe
Mean Squared Error (MSE) is calculated using Equation.
Normalized iris image is used for features extraction.
2.4. The trainee image with the least MSE is declared as the
Overall feature extraction processing is as following:
identified user.
X =X1+X2+...+X5/5
2.5. Repeat steps 2.3 and 2.4 decreasing the value of M
1) First calculate the average 5
gradually from 128 to 1 and record the error obtained in user
2) Calculate cumulative sum from 0: S0 = 0
identification for every fraction of the original feature vector.
3) Calculate the other cumulative sums by adding the difference
2.4 Classification (Matching)
between current value and the average to the previous sum,
Classification is the problem of identifying which of the set
i.e., Si = Si ± l + (Xi -X) for i = 1,2,..., 5. (2)
of categories; a new observation belongs on the basis of a
After calculation cumulative sums, iris codes are
training set of data containing observation whose category
generated for each cells using following algorithm after
membership is known. For the purpose of matching or
obtaining MAX and MIN values among cumulative sums.
classification, various methods are used like hamming
if Si located between MAX and MIN index distance, Weighted Euclidean distance, Normalized
if Si on upward slope set cell's iris code to "1" correlation , support vector Machine(SVM) , artificial Neural
if S5 on downward slope x set Network(ANN, ANFIS, False Accepted / Rejected
cell's iris code to "2" Rate(FAR/ FRR).
else In this paper, hybrid classifiers i.e. combination of
set cell's iris-code to "0" ANN and FAR/ FRR are used for classification/ matching to
This algorithm generates iris codes by analyzing the changes identify individuals identity based on iris code.
of grey values of iris patterns. Upward slope of cumulative
Accepted
sums means that iris pattern may change from darkness to
Feature
brightness. Downward slop of cumulative sums means the ANN
FAR /
FRR
opposite change of upward slope. Vector Rejected
2.3.3 Feature extraction using Haar Transform:
Fig.3. Structure of Hybrid Classifier
This sequence was proposed in 1909 by Alfréd Haar.
Haar used these functions to give an example of a countable A.FAR / FRR:
ortho normal system for the space of square-integral FAR(False Acceptance Rate): The probability that the system
functions on the real line. The study of wavelets, and even incorrectly matches the input pattern to a non- matching
the term "wavelet", did not come until much later. The Haar template in the database. It measures the percent of invalid
wavelet is also the simplest possible wavelet. The technical input which are incorrectly accepted. In case of similarity
disadvantage of the Haar is that it is not continuous, and scale, if the person is imposter in real but the matching score
therefore not differentiable. This property can, however, be is higher than threshold and he is treated as genuine that
an advantage for the analysis of signals with sudden increases the FAR and hence the performance also depends
transitions such as monitoring of tool failure in machines. upon the selection of threshold value.
FRR (False Rejection Rate): The probability that the system
2.3.4 Feature extraction using Hybrid Algorithm: fails to detect a match between the input pattern and a
By studying above algorithm we have used here the matching template in the database. It measures the percent of
Combination of Haar transform and Block sum algorithm. valid inputs which are incorrectly rejected.
The algorithm that we have used for our study on iris
Recognition is as given below: B. Artificial Neural Network (ANN):
1. Creation of feature vector database ANN is a mathematical or computational model that is
1.1. Read the database image. inspired by the structure and function aspects of biological
1.2. Extract the Red, Green and Blue component of that neural networks. A neural network is a system of parallel
image. processors connected together as a directed graph. Each
1.3. Apply Haar transform and Block sum algorithm the neurons of the network is represented as node. ANN is
Red, Green and Blue components of the image. This is the composed of input layers, hidden layers and output layer.
Feature Vector (FV) of that image. ANN has to compare normalized image with original image
1.4. Repeat steps 1 through 3 for every database image. and identify the individual from the image. In this paper, a
2. Testing phase feed forward neural network using Feed Forward Back
2.1. Read the Query image. propagation (FFBP) algorithm is used for iris pattern
2.2. Repeat step 1.2 and 1.3 for the query image so as to classification. The structure of BPNN is shown in fig.
obtain its Feature Vector.
3261
Adjust weight Adjust Weight
Link weight
99
98
98 97.67
97 96.3
96
Output layer
95
95
Hidden layer
Forward Calculation Process
94
93
Fig.4. Structure Of BP Neural Network Haar PCA Block Sum Hybrid
After activation of neural network, the back propagation % in Accuracy
learning algorithm is applied for training.
Fig.5. Accuracy of Haar, PCA, Block Sum and Proposed algorithm
III. EXPERIMENTAL RESULTS AND COMPARISON:
Evaluating the performance of biometric algorithms is a 6
difficult issue. For the purpose of comparison; we implement
these methods according to the published papers. To compare 5
their performance, the version the Chinese Academy of 4
Science Institute of Automation (CASIA) version eye image
database is used in this experiment. CASIA VI Iris Database 3
contains 280 eye images from 28 individuals and every FAR
person has 10 images of eye. All experiments were 2
FRR
performed by using MATLAB version R2010b on core
processor. We use the usual method to locate and normalize 1
iris regions and use the combination of three methods 0
mentioned above to extract the feature. Therefore we only
analyze and compare the accuracy and computational Haar PCA Block Hybrid
complexity of feature extraction. After feature extraction, we Sum
use hybrid classifier for matching stage. i.e. ANN and FAR/
FRR are used for evaluating the result. ANN use back Fig.6. FAR/ FRR in Percentage (%) of Haar, PCA, Block Sum and Hybrid
propagation neural network for classification of iris pattern algorithm
Feature vectors of 10 samples of 28 persons are transmitted
to ANN for classification of iris pattern. ANN randomly
selects the testing data. ANN compares both data. After IV. CONCLUSION:
comparison result is evaluated by FAR/ FRR. FAR/ FRR is
decided either image is accepted or rejected. By using the In this paper, we have discussed feature extraction of
hybrid classifier, the recognition rate is 98%. The table shows iris recognition using Haar transform, PCA, Block sum
the recognition rate shown in Table 1. algorithm with hybrid algorithm. We have applied these
transforms on the iris images for finding out the recognition
Table I. EXPERIMENTAL RESULTS rate and accuracy. Results of this experiment have shown
Methods FAR/ Overall Accuracy that the accuracy in recognition using hybrid algorithm is
FRR using ANN better than block sum, PCA and Haar transform. Also Hybrid
classifier i.e. combination of ANN (Artificial Neural
Haar Transform 5/1 95% Network) and FAR / FRR are used for matching either image
PCA ¾ 96.3% is accepted or rejected. FAR and FRR in percentage with
respective various methods as shown in above graph. Thus
Block Sum 2.43/3.17 97.67% proposed algorithm provides better accuracy and recognition
Algorithm
rate than comparative algorithms.
Hybrid algorithm 5/4 98%
3262
V. Acknowledgment [14] Emmanuel Raj.M.Chirchi, Dr. R. D. Kharadkar, “ Improvement of iris
Pattern authentication system using cumulative sum algorithm” IJEIT
The authors wish to acknowledge Dr. J. Daugman for Volume 2, Issue 5, November 2012 pp 255- 260.
providing iris images and would like to thank Dr. P. Flynn
CASIA (China) for providing the iris databases used in this [15] K. Saminathan, M. Chithra Devi, T. Chakravarthy, “Pair of iris
Recognition for personal identification using Artificial Neural
paper. The authors would also like to thank the reviewers and Networks” IJCSI International Journal of Computer Science Issues,
editors for providing their feedback and useful suggestions. Vol. 9, Issue 1 No 3 January 2012 pp 324 – 327.
[16] Anjana Peter, Revathi N, Ms. Merlin Mercy, “ Neural network based
Matching approach for Iris recognition” IARCET volume 2, Issue 2,
February 2013 pp 618 – 624.
References
[17] Shivni Godr, Ravindra Panchariya, “ Comparison of different Neural
Networks for iris recognition” International Journal of Engg Science &
[1] Sudeep D. Thepade, Pooja V. Bidwai, “ Contemplation of image based Advanced Technology (IJESAT) Volume 2, Issue – 4, pp 1168- 1172.
UIris recognition”, International journal of Engg. Research and
application ISSN- 2248- 9622, Volume 3, Issue 2, March- April 2013,
pp 1056-1066.
[2] H.eba Mohamed, Abdul Hamid, a thesis of ,” Iris Verification
System”.
[3] Pravin S. Patil, S. R. Kolhe, R. V. Patil,“ Performance evaluation in
iris recognition and CBIR system based on Phase Congruency”,
International journal of computer application(0975- 888) Volume 47-
No.14, June 2012.
[4] Asim Ali Khan, Smriti Kumar and Mister Khan, “ Iris pattern
Recognition using Support Vector machines and Artificial neural
networks” , international journal of innovative research in electrical
electronics instrumentation and control engg. Vol.2 issue 12 December
2014.pp 2208- 2212.
[5] Muhammad Waheed, a thesis on “ iris recognition using image
processing and neural network”, Nicosia, 2014.
[6] Sudha Gupta, Viral Doshi, “ iris recognition system using biometric
template matching technology”, 2010 international journalof computer
applications (0975- 8887), Volume 1- No.2pp 1-4.
[7] Aparna Gale, S. S. Salankar, “ Performance analysis on iris feature
extraction using PCA, Haar transform and block sum algorithm”,
international journal of Engg and advanced tecnology(IJEAT), ISSN:
2249- 8958, Volume. 4, April 2015, pp 46- 49.
[8] Priyanka Mukherjee, ankita dutta , Pranab Das, “ Recent developments
in iris based Biometric Authentication Systems”. IJRET, eISSN: 2319-
1163, volume 02 issue . 11 Nov. 2013.pp 266- 270.
[9] Dr. Sanjay Ganorkar, Mirza Shuaur Rahman, “ recognition system
based on Neural Networks”. IJSER, volume 4, issue 12, December
2013, pp 847- 850.
[10] Mayank Vasta, Richa Singh, Afzel Noore, “ Improving Iris
recognition and performace using segmentation, quality enhancement,
match score fusion and Indexing”. 1083-4419/$25.00 © 2008 IEEE,
pp 1-15.
[11] Arun Ross, West Virginia University, “ Biometric, Iris recognition: the
path forward” published by the IEEE computer society 0018-
9162/10/$26.00 @2010 IEEE pp 30 – 36.
[12] Mam Nabti, Lahouari Ghouti and Ahmed Bouridane, “ An effective
and Fast iris recognition system based on a combined multiscale
feature extraction technique”. 0031-3203/$30.00 _ 2007 Pattern
Recognition Society. Published by Elsevier Ltd. pp 868 -880.
[13] Richard P. Wildas, ‘‘Iris recognition: an emerging biometric
Technology’’ 0018–9219/97$10.00 ¤ 1997 IEEE VOL. 85, NO. 9,
September 1997 pp 1348- 1363.
3263
