A COST-EFFECTIVE IRIS RECOGNITION SYSTEM
USING LINEAR DISCRIMINANT ANALYSIS
AND CROSS-CORRELATION TECHNIQUES
Dr. A K Ramani
Department of computer Science, DAVV, Indore, MP, India
Prof. Sanjay Silakari
Department of computer Science, SATI, Vidisha, MP, India
Pinaki A Ghosh
Department of computer Science, SATI, Vidisha, MP, India
Keywords: biometrics, iris r
ecognition, user authentication, linear discriminant analysis, discrimination.
Abstract: Authorization and identification has become a vital part of security systems of any society. With the
changing of technology implementations in the present scenario, every country specially developing
countries like India needs a cost-effective and reliable solution for authentication system. In this paper,
efficient technique for iris recognition system is described which provides a reliable authentication at low
cost. The proposed system uses linear discriminant analysis and cross correlation methods for identification
and verification purpose. The system was implemented and tested using a dataset of 80 samples of iris with
different contrast quality. The classification rate compared with the well-known methods is also discussed.
1 INTRODUCTION
Biometric is automated methods of identifying a
person or verifying the identity of a person based on
a physiological or behavioral characteristic (Eric J.
Lerner, Feb 2000)(K. L. Kroeker, 2002)(S. Liu et al.,
2002). Examples of physiological characteristics are
fingerprint, hand geometry, facial characteristics and
iris recognition. Behavioral characteristics include
traits that are learned or acquired. Dynamic
signature verification, speaker verification etc. are
examples of behavioral characteristics.
In any pattern recognition system, the main issue
is th
e relation between inter-class and intra-class
variability. For example in face recognition the
difficulty is that the face is changeable organ. It
displays a number of expressions and it varies with
viewing angle, pose and age. it has been proved that
for facial recognition the images taken a year apart,
the best algorithms have error rates of 43% (Phillips
2000) to 50% (Pentland 2000). The inter-class
variability is limited because different faces possess
the same basic set of features. For these reasons iris
patterns become a good alternative approach to
reliable visual recognition (K. L. Kroeker, 2002) of
persons. The iris has the great mathematical
advantage that its pattern variability among different
persons is enormous. As an internal organ of the eye,
the iris is well protected from environment and
stable over time. The iris begins to form in the third
month of gestation and the structure creating its
pattern are complete by the eighth month. Its
complex pattern can contain many distinctive
features such as arching
ligaments, ridges, furrows,
crypts, rings, corona, freckles and a zigzag
collarette. The density of melanin pigment
determines Iris color. Blue irises resulting from an
absence of pigment. The properties of the iris that
enhance its suitability for use in biometric
identification include:
Protecte
d from the external environment
Im
possibility of surgically modifying without
the risk of vision
140
Silakari S., K. Ramani A. and A. Ghosh P. (2004).
A COST-EFFECTIVE IRIS RECOGNITION SYSTEM USING LINEAR DISCRIMINANT ANALYSIS AND CROSS-CORRELATION TECHNIQUES.
In Proceedings of the First International Conference on E-Business and Telecommunication Networks, pages 140-145
DOI: 10.5220/0001387001400145
Copyright
c
SciTePress
Physiological response to light
Ease of registering its image at some distance
2 IRIS FEATURE AND PROCESS
There are verities of features that can be used to
distinguish one iris from another. One of the main
characteristics is the trabecular meshwork, a tissue
which gives the appearance of dividing the iris in a
radial fashion that is permanently formed by the
eighth month of gestation. During the development
of iris there is no genetic influence on it, a process
known as chaotic morphogenesis that occurs during
the seventh month of gestation, for which identical
twins have four different irises. The iris is protected
behind the eyelid and cornea unlike other biometrics
such as fingerprints or face, the chances of damage
is minimal or nil. The iris is not a subject to be
change with age, it means its pattern remain stable
till death. Generally, the process of iris recognition
system includes the following steps:
Image Acquisition
Iris Localization
Image Optimization
Comparison
The image of iris can be captured using a
standard camera using both visible and infrared light
and may be either a manual or automated procedure.
In manual procedure, the user needs to adjust the
camera to get the iris in focus and needs proper user
training to be successful. The automated process
uses a set of cameras that locate the face and iris
automatically. Once the camera has located the eye,
the image is thenanalyzed to identify the outer
boundary of the iris, the pupillary boundary and the
center of pupil. This information is used to produce
a vector record called iriscode. This record is then
stored into a database for future comparison.
3 RELATED WORK
Daugman’s work: the visible texture of a person’s
iris in a real-time video image is encoded into a
compact sequence of multi-scale quadrature 2-D
Gabor wavelet coefficients, whose most-significant
bits comprise a 256-byte iris code (J. Daugman,
“How Iris Recognition Works”)(J. Daugman, Nov
1993)(J. Daugman, Dec 2001).
Wildes’ work: is very similar to the above-
mentioned method. A laplacian pyramid is used to
apply a 2-D transformation. A match value is
calculated for the four bands through spatial
correlation (R. Wildes, Sept 1997).
Boles’ work: based on calculating the zero-
crossings representations of the wavelet transform.
These representations are stored as templates and are
used for the matching algorithm (W. Boles, 1997).
4 THE PROPOSED SYSTEM
This system is divided into these parts:
A. Image Acquisition: This is the stage of acquiring
the eye image using digital camera.
B. Iris Localization: Finding the boundary between
the pupil and the iris, the outer boundary of iris and
the center of the pupil.
C. Polar Transformation: Using the center and the
radius we find a polar coordinate system. In this
system the feature of the iris is extracted.
D. Identification: Using linear discriminant analysis
we found a match for the acquired iris feature.
E. Verification: Using cross correlation method it
verifies the identified image to get efficient search
result.
A. Image Acquisition
The iris is a relatively small (1 cm diameter),
dark object and that human operator are very
sensitive about their eyes, this matter requires
careful engineering. The following points should be
concern:
Describe to acquire images of the iris with
sufficient resolution and sharpness to support
recognition
It is important to have good contrast in the
interior iris pattern without resorting to a level of
Illumination that annoys the operator
The image should be well framed (i.e. centered)
Noises in the acquired images should be
eliminated as much as possible
A COST-EFFECTIVE IRIS RECOGNITION SYSTEM USING LINEAR DISCRIMINANT ANALYSIS AND
CROSS-CORRELATION TECHNIQUES
141
B. Iris Localization
Since the value of the pixels in the pupil not
always be zero so we need an edge detection
algorithm (M. Turhan, Apr 1999) to make all values
of the pupil to be zero to easy determination of the
pupil center and then get the pupil boundary. Get the
center of the pupil by counting the number of black
pixels of each column and row. Then get each row
and column that has maximum number of pixels.
Then determine the center as:
Get the position of first and last pixels (X
a, Ya) and
(X
b, Yb) of this row. Find the center X0 = (Xa + Xb) /
2. Similarly apply same step to find Y
0 = (Ya + Yb) /
2. Consequently, the radius of the virtual circle of
the pupil can be determined. A similar procedure is
extended to locate the outer boundary that can be
apparent by using the mid-point algorithms of circle.
C. Polar Transformation
The image should be transformed into polar
coordinate system. The q(q Î[0;2p]) parameter and
dimensionless p ( p Î[0;1]) parameter describe the
polar coordinate system. Thus the following
equations implement:
I(x(p, q), y(p, q)) à I(p, q)
Where,
x(p, q) = (1-p)*x
p(q) + p*xi(q)
y(p, q) = (1-p)*y
p(q) + p*yi(q)
x
p(q) = xp0(q) + rp*cosq
y
p(q) = yp0(q) + rp*sinq
x
i(q) = xi0(q) + ri*cosq
y
i(q) = yi0(q) + ri*sinq
Where r
p and ri are respectively the radius of the
pupil and the iris. Figure-4 shows how iris image is
converted to polar coordinates.
iris region
unwrapped image after transformation
D. Identification
Linear Discriminant Analysis (LDA), is perhaps
the best known technique for classifying an
observation of unknown class membership into one
of two populations on the basis of predictors. It
happens frequently in literatures that “classification
and “discrimination” are used interchangeably,
though the former is used sometimes when the
problem is one of “clustering” (P. N. Belhumeur et
al, Jul 1997)(J. Buckheit et al.).
Given C groups or classes {X
1, X2, ... , Xc} contains
N sample images {x
1, x2, ... , xn} such that each
image belongs to exactly one of the C classes. If the
total scatter matrix S
T is defined as:
N
ST = (Xk - µ)(Xk - µ)T
k-1
Where n is the number of sample images, and m Î Rn
is the mean image of all samples, then after applying
the linear transformation W
T, the scatter of the
transformed feature vectors {y
1, y2, ... , yn} is
W
TSTW. In PCA, the projection Wopt is chosen to
maximize the determinant of the total scatter matrix
of the projected samples, i.e.,
W
opt = arg max WTSTW
= [W
1 W2 ... Wm]
ICETE 2004 - SECURITY AND RELIABILITY IN INFORMATION SYSTEMS AND NETWORKS
142
Where {Wi i = 1, 2, ..., m} is the set of n-
dimensional eigen-vectors of S
T corresponding to the
m largest eigenvalues.
Linear Discriminant Analysis (LDA) is an
example of a class specific method, in the sense that
it tries to “shape” the scatter in order to make it more
reliable for classification. This method selects W in
such a way that the ratio of the between-class scatter
and the within-class scatter is maximized. Let the
between-class scatter matrix be defined as
c
SB = Ni (µi - µ)(µi - µ)T
i=1
and the within-class scatter matrix be defined as
c
SW = ∑ ∑ N i(Xk - µi)(Xk - µi)T
i=1 XkXi
Where m
i is the mean image of class Xi , and Ni is the
number of samples in class X
i. If SW is nonsingular,
the optimal projection Wopt is chosen as the matrix
with orthonormal columns, which maximizes the
ratio of the determinant of the between-class scatter
matrix of the projected samples to the determinant of
the within-class scatter matrix of the projected
samples, i.e.,
W
TSBW
W
opt = arg max ⎯⎯⎯⎯⎯
W
TSWW
where {W
i i =1, 2, ... , m} is the set of generalized
eigen-vectors of S
B and SW corresponding to the m
largest generalized eigenvalues {l
i i = 1, 2, ... , m}
i.e., S
BWi = λiSWWi i = 1, 2, ... , m
Note that there are at most C - 1 nonzero generalized
eigen-values, and so an upper bound on m is C - 1,
where C is the number of classes. The solution is
given by the min(C - 1) eigen-vectors (called
“canonical variates” by Rao) of S
W—1SB.
E. Verification
Once the iris was identified the system verifies
it using cross correlation technique to produce better
search results.
Given two images 1 and 2, image-1 is the template
image and image-2 is the image to be searched
∑ ∑ (g
1(x, y) - µ1) (g2(x, y) - µ2)
x y
ρ = ⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯
√ ∑ ∑ (g
1(x, y) - µ1)2 ∑ ∑ (g2(x, y) - µ2)2
x y x y
-1 ≤ ρ ≤ 1
where,
g
1(x, y) = individual gray values of template matrix
µ
1 = average gray value of template matrix
g
2(x, y) = individual gray values of search matrix
µ2 = average gray value of search matrix
5 RESULTS
To show the effectiveness of the proposed system, in
our implementation, we used 10 classes each contain
5-9 iris images acquired under different conditions.
The images were collected as gray-scale images. For
each iris class, we randomly choose four samples for
training and the rest for testing. In identification
tests, an average correct classification rate of
98.97% is achieved. These tests were performed on
300 MHz uniprocessor Pentium based system.
Operating states of our previous system.
Operating states of system using LDA.
Operating states of the proposed system.
Comparison with Existing Methods
Daugman documented first iris recognition
system in 1993 (J. Daugman, Nov 1993). This
system is based on wavelet transformation. It is
implemented on a RISC general-purpose CPU and
32X32 array of 74F86 ICs. Wildes described a
system for personal verification based on automatic
iris recognition in 1996 (R. Wildes, Sept 1997). This
system used a very similar technique l i k e
Daugman’s. It was implemented on Sun
SPARCstation 20 and written in UNIX C Shell
languages without optimization. Boles proposed a
system using zero-crossing representation (W.
Boles, 1997). Our early works was based on
correlation method and PCA. All these algorithms
A COST-EFFECTIVE IRIS RECOGNITION SYSTEM USING LINEAR DISCRIMINANT ANALYSIS AND
CROSS-CORRELATION TECHNIQUES
143
are based on gray image. A gray iris image can
provide enough information to identify different
individuals. The methods proposed by Daugman and
Wildes are the best among all, but these methods
require expensive hardware like RISC & parallel
processing architecture systems to implement it. The
proposed system can run on uniprocessor system. So
we can say that this proposed system is very much
cost-effective between these methods.
6 CONCLUSION
In this paper, we have presented a new and effective
algorithm for iris recognition. The proposed system
uses linear discriminant analysis for identification
then verifies the result using cross-correlation.
Experimental results have shown that the proposed
algorithm achieves
high performance at low cost. Developing
countries like India where population density and
crime rates are increasing day by day, such high
technology oriented authentication systems are
desperately in requirement. The existing iris
recognition systems require high cost computing
system, which is not feasible for developing
countries. The proposed system is highly suitable for
such countries and can be used for a wide range of
application areas involving forensic science, crime
search, driver’s license, passport, etc.
Some efficient techniques are used in this
proposed method, these are:
A computer graphics algorithm for detecting the
center of the pupil and localizing the iris area.
Transforming the localized iris area into a
simple coordinate system.
Identification process based on LDA.
ICETE 2004 - SECURITY AND RELIABILITY IN INFORMATION SYSTEMS AND NETWORKS
144
REFERENCES
Eric J. Lerner, “Biometric Identification” The
Industrial Physicist, American Institute of Physics, Feb
2000.
K. L. Kroeker, “Graphics and Security: Exploring Visual
Biometrics” Computer Graphics and Applications,
IEEE Computer Society, 2002.
S. Liu, M. Silverman, “A Practical Guide to
Biometric Security Technology” Security, IEEE
Computer Society, 2002.
J. Daugman, “How Iris Recognition Works”.
J. Daugman, “High Confidence Visual Recognition of
Persons by a test of Statistical Independence” IEEE
Trans. Pattern Analysis and Machine Intelligence, vol.
15, no. 11, Nov 1993.
J. Daugman, “The Importance of being Random:
Statistical Principles of Iris Recognition” The Journal
of the Pattern Recognition Society, Dec 2001.
R. Wildes, “Iris Recognition: An Emerging Biometric
Technology” Proc. of IEEE, vol. 85, no. 9, Sept 1997.
W. Boles, “A Security system based on Human Iris
Identification using Wavelet Transform” Proc. Of the
first International Conference on Knowledgebased
Intelligent Electronic System, 1997.
M. Negin, Thomas A. Chmielewski Jr. “An Iris Biometric
System for Public and Personal Use” IEEE, Feb 2000.
M. Leonidou, “Iris Recognition: Closer than we Think?”.
M. Turhan, “Multi Component Edge Detection Algorithm
Revised Apr 1999.
M. Lyons, J. Budynek, A. Plante, S. Akamatsu,
Classifying Facial Attributes using a 2-D Gabor
Wavelet Representation and Discriminant Analysis
Proc. of 4th International Conf. of IEEE Computer
Society, March 2000.
P. N. Belhumeur, J. P. Hespanha, D. J. Kriegman,
Eigenfaces vs. Fisherfaces: Recognition using Class
Specific Linear Projection” IEEE Trans. On pattern
Analysis and Machine Intelligence, vol. 19, no. 7, July
1997.
J. Buckheit, D. Donoho, “Improved Linear Discrimination
using Time Frequency Dictionaries”.
A COST-EFFECTIVE IRIS RECOGNITION SYSTEM USING LINEAR DISCRIMINANT ANALYSIS AND
CROSS-CORRELATION TECHNIQUES
145