Iris Recognition System With Frequency Domain Features Optimized With PCA and SVM Classifier
Iris Recognition System With Frequency Domain Features Optimized With PCA and SVM Classifier
172
Abstract
Applications such as immigration control, aviation security, bank
and other financial transactions, access to defence organization
requires a more reliable and authentic identification system. Iris
is now considered to be one of the most time invariable biometric
features of a person for recognition. Several iris recognition
techniques were proposed with considerable focus on improving
the false acceptance rate and minimizing false rejection rate.
Most of the proposed techniques are tested with Mat lab and not
keeping the detection and recognition time in mind. In this work
we propose a novel iris recognition system with iris localization
to segment and recognize color iris with highest speed and
accuracy. Custom software for iris image processing is developed
in C#.Net (.Net 3.5). Frequency domain magnitude and phase
features are used for image feature representation. Support vector
machines with winner takes it all configuration are used for
classification. Tests shows 97% accuracy with average time of 31
milliseconds seconds for classifying each test image.
Keywords: SVM, iris Recognition, PCA, FFT, C#.Net.
1. Introduction
Personal identification is gaining more importance owing
to increased security requirement of modern day man.
Some of the few methods for personal identification and
verification are token based method which makes use of
ID cards for identification/authorization and memory
based method which requires the user to know passwords
for identification purposes. The disadvantage of these
traditional methods is that they are not reliable as the
Copyright (c) 2012 International Journal of Computer Science Issues. All Rights Reserved.
IJCSI International Journal of Computer Science Issues, Vol. 9, Issue 5, No 1, September 2012
ISSN (Online): 1694-0814
www.IJCSI.org
2. Related work
Zhonghua Lin and Bibo Lu [1] suggested the iris
recognition method based on the optimized Gabor filter.
The iris image was preprocessed and normalized to reduce
it into 512 columns by 64 rows of rectangular image. The
image was segmented according to the parameters of Gabor
filter and edge response of different directions was
obtained. The feature vector was created using iris code and
Hamming distance method was used for recognition and
matching. Zhang Shunli et al., [2] proposed the iris feature
extraction and recognition algorithm based on empirical
mode decomposition. The iris signal was decomposed into
several Intrinsic Mode functions using Empirical Mode
Decomposition. Among n intrinsic mode functions only the
best performing intrinsic mode functions were selected to
form feature vector. Hamming distance was used for
matching. Zhong hua Lin [3] proposed the iris recognition
method based on the Natural Open eyes. The noisy features
like eyelid, eyelashes were removed using edge directional
operator, canny edge operator and voting mechanism. The
iris feature points were obtained using directional
information, length information, width information of
texture to form feature vector. Matching was performed
using pattern matching methods. Karen P Hollingsworth et
al., [4] proposed the algorithm of Iris recognition using
fusion of Hamming Distance and Fragile bit distance. All
the bits in iris code are not very consistent. There will be
some bits called as fragile bits which change their value for
different images of the same iris. The location of the fragile
bit tends to be consistent for the same iris. In the algorithm
the fragile bit distance of two iris codes was taken as metric
to define the feature of the iris. The scores of the fragile bit
distance and Hamming distance were fused to generate the
feature vector for classification. Christian Rathgeb and
Andreas Uhi [5] proposed the method of bit reliability
driven template matching for iris recognition. During
feature extraction phase the texture was divided into strips
to obtain ten, one dimensional signals. A dyadic wavelet
transform was applied to get twenty sub-bands. From each
sub-band using suitable threshold bit code was generated. In
the second phase Log Gabor function was applied on a row
wise convolution pattern to get another feature vector.
Using reliability mask binary iris code was formed.
Matching was performed using Hamming distance code.
Nawal Alioua et. al., [6] presented a method for eye state
analysis using iris detection based on circular Hough
transform. The Circular Hough transform is applied to
segregate the eye from the image of the face. The eye state
analysis using Circular Hough transform was compared
with the other methods like prewitt, Sobel, Log and canny
edge operators. Ali Abd Almisreb et al., [7] suggested
enhancement pupil isolation method in iris recognition. The
method involves converting the original pupil into white
using morphological process. Thus the exact dimension of
173
Copyright (c) 2012 International Journal of Computer Science Issues. All Rights Reserved.
IJCSI International Journal of Computer Science Issues, Vol. 9, Issue 5, No 1, September 2012
ISSN (Online): 1694-0814
www.IJCSI.org
174
3. Contribution
The proposed iris recognition system is implemented with
C#.Net which is considered as fastest processing language
after C++. The C#.Net language is selected keeping in
mind of the type of applications that the system may get
integrated with .Net platform provides a real time interface
for connecting to networks through web services and XML
protocol. All the media types are serializable as XML data.
Therefore the developed system acts as a test bed to test
various Iris recognition techniques over the network. The
contribution of the paper is also significant in designing
and modeling support vector machines of two classes in
C# and extending it for a multiclass problem. The
segmentation algorithm is developed for color images and
fast segmentation is possible with any user input.
Organization: The paper is organized into the following
sections. Section 2 is an overview of related work. The iris
recognition model is described in Section 4. Section 5 is
the algorithm of iris recognition system with methodology
and feature extraction. Classification is presented in
Section 6 and Conclusion is contained in Section 7.
Copyright (c) 2012 International Journal of Computer Science Issues. All Rights Reserved.
IJCSI International Journal of Computer Science Issues, Vol. 9, Issue 5, No 1, September 2012
ISSN (Online): 1694-0814
www.IJCSI.org
6.
4. Model:
Test image
Preprocesses
Preprocesses
Segmentation
Segmentation
2D-FFT
2D-FFT
PCA
PCA
5. Methodology
5.1 Segmentation
During image acquisition an iris image has pupil and outer
retina as unnecessary parts in the image. If not segmented
properly, then the feature extraction process takes into
account the properties of these parts also. This reduces the
overall recognition efficiency. The proposed work
emphasizes on fast processing of the data. Therefore we
developed a novel Iris segmentation as proposed below.
1.
2.
3.
4.
5.
175
Fig 2a. Sample Iris Image and the Step wise Segmentation Result
Copyright (c) 2012 International Journal of Computer Science Issues. All Rights Reserved.
IJCSI International Journal of Computer Science Issues, Vol. 9, Issue 5, No 1, September 2012
ISSN (Online): 1694-0814
www.IJCSI.org
176
Fig 2b. Sample Iris Image and the Step wise Segmentation Result
Fig 3. Two samples of iris images with their Phase and Frequency Plots
that most iris images are circular and the central iris has
most invariability. The actual iris pattern information will
be present outside the central block of values. Iris images
will be symmetric around the center. Following steps are
performed to reduce 256x256 features describing the iris
pattern.
1.
2.
Series1
Series2
Series3
Series4
Fig 4 shows the Magnitude plot of 7 iris images from the database
where FFT is carried out on scaled images of size 128x128.
3.
4.
1
17
33
49
65
81
97
113
Copyright (c) 2012 International Journal of Computer Science Issues. All Rights Reserved.
IJCSI International Journal of Computer Science Issues, Vol. 9, Issue 5, No 1, September 2012
ISSN (Online): 1694-0814
www.IJCSI.org
177
0.025
0.02
0.015
0.01
0.005
0
0
10
20
30
40
50
60
70
6.
7.
(4)
6. Classification
6.1 Support Vector Machines
Support Vector Machines (SVMs) is a supervised learning
method which can be used for both classification and
k=1;
do
Create an initial SVM problem as AxB where A=k,
b={k+1,k+2,.n}
Copyright (c) 2012 International Journal of Computer Science Issues. All Rights Reserved.
IJCSI International Journal of Computer Science Issues, Vol. 9, Issue 5, No 1, September 2012
ISSN (Online): 1694-0814
www.IJCSI.org
Results
Classification Rate
178
2500
2000
1500
1000
500
0
gabor with 4
filter banks
15 30 50 64
Gabor with 8
14 100 250 432 731 18002300
filter banks
Proposed
Technique
2
15 30 50 64
GABOR
CONVOLVE
WITH 4 100 100 99 99 97 97 95
FILTER
BANKS
GABOR
CONVOLVE
WITH 8 100 100 99 99 98 97.5 97
FILTER
BANKS
PROPOSED 100 100 99 99 99 96 96.5
Fig 7. Number of Classes v/s Rate of Correct classification comparison
with proposed technique, Gabor based filter bank features and SVM
classifier.
7. Conclusion
The performance gain in terms of classification accuracy
compared with classification time is non linear. Increasing
the accuracy by a percent, then processing time is also
increased several folds. This is avoided in the proposed
technique. One of the interesting facts that come up is that
when number of person classes is below 15, the efficiency
of the proposed technique is better than that of
conventional Gabor filter based technique. Though for any
practical applications, number of classes may be huge.
Higher accuracy can be achieved by either increasing
number of samples per class in the training phase or
combining a hybrid approach of Gabor convolve and
proposed technique such that the Gabor features are
extracted and considered for higher number of classes in
series with proposed frequency domain features optimized
by principal components.
Copyright (c) 2012 International Journal of Computer Science Issues. All Rights Reserved.
IJCSI International Journal of Computer Science Issues, Vol. 9, Issue 5, No 1, September 2012
ISSN (Online): 1694-0814
www.IJCSI.org
References:
[1] Zhonghua Lin and BiBo Lu, Iris Recognition Method Based
on the Optimised Gabor Filter, Third IEEE International
Congress on Image and Signal Processing, 2010,
pp 1868-1872.
[2] Zhang Shunli, Han Min, Sun Weifeng, Yang Mingqiang,
Iris Feature Extraction and Recognition Based on Empirical
Mode Decomposition, Seventh International Conference on
Fuzzy Systems and Knowledge Discovery, 2010,
pp 2633-2636.
[3] Zhongua Lin A Novel Iris Recognition Method Based on
the Natural Open Eyes, Tenth IEEE International
Conference on Signal Processing, 2010, pp 1090-1093.
[4] Karen P Hollingsworth, Kavin W Bowyer and Patrick J
Flynn, Improved Iris Recognition Through Fusion of
Hamming Distance and Fragile Bit Distance, IEEE
Transaction on Pattern Analysis and Machine Intelligence,
Vol. 33, No. 12, 2011, pp 2465 - 2476.
[5] Christian Rathgeb and Andreas Uhi, Bit Reliability Driven
Template Matching in Iris Recognition, Fourth IEEE Pacific
Symposium on Image and Video Technology, 2010,
pp 70-75.
[6] Nawal Alioua, Aouatif Amine, Mohammed Rziza, Driss
Aboutajdine, Eye State Analysis using Iris Detection based
on Circular Hough Transform IEEE International
Conference on Multimedia Computing and Systems, 2011,
pp 1-5.
[7] Ali Abul Almisreb, Nooritawati Mohammad Jahir, Adzrool
Idzwan Ismail and Ramli Abdullah, Enhancement Pupil
Isolation Method in Iris Recognition, IEEE International
Conference on System Engineering and Technology, 2011,
pp 1-4.
[8] Fitri Arnia, Nuriza Pramita, Enhancement of Iris
Recognition System based on Phase Only Correlation,
Telkomnika , Vol. 9, No. 2, 2011, pp 387-394.
[9] Sambita Dalal and Tapasmini Sahoo, A selective Feature
Matching Approach for Iris Recognition, International
Journal of Computer Applications (975-8897), Vol. 41,
No. 20, March 2012, pp 34-39.
[10] P Radu, K Sirlantzis, W G J Howells, F Deravi and
S Hoque, Information Fusion for Unconstrained Iris
Recognition, International journal of Hybrid Information
Technology, Vol. 4, No. 4, October 2011, pp 1-12.
[11] Bhavana Chouhan and Shailaja Shukla, Iris Recognition
System using Canny Edge Detection for Biometric
Identification, International Journal of Engineering Science
and Technology (ISSN: 0975-5462) Vol. 3, No. 1, January
2011, pp 31-35.
[12] M.Z.Rashad, M Y Shams, O Nomir and R M El-Awady,
Iris Recognition based on LBP and Combined LVQ
Classifier, International Journal of Computer Science and
Information Technology, Vol. 3, No. 5, October 2011,
pp 67-78.
[13] P Manikandan and M.Sundarrajan, Discrete Wavelet
Features Extraction for Iris Recognition based Biometric
Security, International Journal of Electronic Engineering
Research (ISSN 0975-6450), Vol. 2, No. 2, 2010,
pp 237-241.
[14] Hongying Gu, Shunguo Qiao and Cheng Yang, An
Efficient Iris Localization Algorithm Based on Standard
[15]
[16]
[17]
[18]
179
Copyright (c) 2012 International Journal of Computer Science Issues. All Rights Reserved.
IJCSI International Journal of Computer Science Issues, Vol. 9, Issue 5, No 1, September 2012
ISSN (Online): 1694-0814
www.IJCSI.org
180
Copyright (c) 2012 International Journal of Computer Science Issues. All Rights Reserved.