[EEE/OSA/[APR Interational Conference on [nfonnatics, Electronics & Vision
Low Complexity Iris Recognition using Curvelet
Transform
Afsana Ahamed
Department of EEE
Bangladesh University of Engineering and Technology
Dhaka, Bangladesh
afsana4@gmail.com
Abstract-In this paper, a low complexity technique is
proposed for iris recognition in the curvelet transform
domain. The proposed method does not require the detection
of outer boundary and decreases unwanted artefacts such as
the eyelid and eyelash. Thus, the time required for
preprocessing of an iris image is signifcantly reduced. The
zero-crossings of the transform coefcients are used to
generate the iris codes. Since only the coefcients from
approximation subbands are used, it reduces the length of the
code. The iris codes are matched employing the correlation
coefcient. Extensive experiments are carried out using a
number of standard databases such as CASIA- V3, UBIRIS.vl
and UPOL. The results reveal that the proposed method using
the curvelet transform provides a very high degree of
accuracy (about 100%) over a wide range of images with a
low equal error rate (EER) and a signifcant reduction in the
computational time, as compared to those of the
state-of-the-art techniques.
Keywords-iris recognition; curvelet transform; correct
recognition rate; equal error rate
1.
I
NTRODUCTION
Biometric recognition systems are becoming popular in
recent times due to their robustness to exteral effects and for
the ability to offer a high degree of uniqueness and stability
for person identifcation [I]. A biometric recognition system
uses a person's physiological and/or behavioral characteristics
as features that include face, fngerprint, palm print, voice
patters, ear, iris and gait. Recently, iris recognition has
emerged as a superior method of identifcation technology.
The human iris is an annular region between the pupil
(generally darkest portion of the eye) and sclera. [t has many
interlacing minute characteristics such as feckles, coronas,
stripes, frrows, crypts and so on. The iris patters of the two
eyes of an individual or those of identical twins are
completely independent and uncorrelated, which is largely
detennined by the pre-natal development. [t is reported that an
iris being similar to another oane is of extremely low
probability, about 1 in 10
72
[2]. [n addition, the iris is visible,
making it highly suitable for practical noninvasive
identifcation processes.
The concept of automated lflS recognition is frst
introduced by Flom and Safr in [2]. Since then, a number of
methods have been proposed in the literature for effective iris
978-1-4673-1154-0112/$3l.00 ©2012 IEEE
Mohammed Imamul Hassan Bhuiyan
Associate Professor, Department of EEE
Bangladesh University of Engineering and Technology
Dhaka, Bangladesh
imamhas@gmail.com
recogmtlOn. [n the pioneering work of Daugman [3,4],
multiscale Gabor flters are used to demodulate texture phase
structure information of the iris. This results in 1024
complex-valued phasors at different scales. Each phasor is
quantized to one of the four quadrants in the complex plane.
The resulting 2048-component iris code is used to describe an
iris. The difference between a pair of iris codes is measured
by their Hamming distance. The method of Daugman
provides high accuracy and is employed in several
commercial iris recognition systems. The drawback of this
method is its rather long length of the iris code considering
the large volume of iris data to be compared. Wildes [5]
represents the iris texture with a Laplacian pyramid
constructed with four different resolution levels and uses the
normalized corelation to determine whether the input image
and the model image are fom the same class. However, it
gives lower accuracy as compared to that of [3] and suffers
fom a high computational complexity. Boles and Boashash
[6] calculate the zero-crossings of an one-dimensional (1 D)
wavelet transform at various resolution levels of a concentric
circle on an iris image to generate the iris code. [n [7], the 4
th
level Haar wavelet coefcients are quantized to form the iris
code, later classifed using a competitive learing neural
network. [n [8], overlapped patches of normalized iris are
subjected to discrete cosine transform (DCT); the differences
of the coefcients are then quantized to fonn the iris code,
later matched by Hamming distance. In [9], phase
components of 2-D Fourier transform coefcients are used as
features that are classifed using the Hamming distance. In
[10], both the phase and magnitude of the Gabor wavelet
outputs are used as features that are matched region-wise
using the Euclidean distance. [n [11], the iris features are
extracted using the oriented separable wavelet transforms,
called directionlets, and compared using a weighted Hamming
distance. In [12], a co-occurrence matrix is generated using
the values of various statistical measures calculated fom the
Contourlet coefcient for matching. In [13], a novel
normalization method is employed for iris localization and the
corresponding Contourlet coefcients are classifed using a
Support Vector Machine (SVM). Note that the methods of
[12] and [13] report recognition rates for a rather limited
number of classes.
Recently, the multi-scale Curvelet transform [14] has
emerged as a highly effective tool for sparse image
representation, much better than the traditional wavelets.
However, to the best of our knowledge very limited work has
been done on curvelet transform-based iris recognition
ICIEV 2012
MATLAB code can be download from matlab1.com
matlab1.com