A Novel Iris Recognition System

Abstract

Iris recognition as an emerging biometric recognition approach and it’s becoming a very active topic in both research and practical applications. The pattern of the human iris differs from person to person, even between monocular twins. This paper proposes a modified iris localization method and normalization method. In the iris localization after invert the iris image edges are detected by using the canny edge detection. The average of the edge points as taken as coarse center of the pupil. Using coarse center of pupil, Circular Hough Transform method is applied to determine the location of the iris boundaries. The normalization is used for converting the circular iris into rectangular form with fixed dimensions. The proposed method is for non-concentric iris. In this paper 2D Log-Gabor Filters are used for feature extraction and hamming distance is used for matching. The experimental result shows the proposed method has an encouraging performance.

Full text

Page 1

A Novel Iris Recognition System


V C Subbarayudu
Institute for Development and Research in
Banking Technology
vcsubbarayudu@mtech.iderbt.ac.in


Abstract

Iris recognition as an emerging biometric
recognition approach and it’s becoming a very
active topic in both research and practical
applications. The pattern of the human iris differs
from person to person, even between monocular
twins. This paper proposes a modified iris
localization method and normalization method. In
the iris localization after invert the iris image edges
are detected by using the canny edge detection. The
average of the edge points as taken as coarse center
of the pupil. Using coarse center of pupil, Circular
Hough Transform method is applied to determine the
location of the iris boundaries. The normalization is
used for converting the circular iris into rectangular
form with fixed dimensions. The proposed method is
for non-concentric iris. In this paper 2D Log-Gabor
Filters are used for feature extraction and hamming
distance is used for matching. The experimental
result shows the proposed method has an
encouraging performance.


1. Introduction

Biometrics is a process of identifying an
individual by his/her physiological or behavioral
characteristics. Biometrics is one of the most
important and reliable methods for computer-aided
personal identification. Its can be used in wide range
of applications, in government programs such as
national ID cards and as well as having personal
applications in areas such as logical and physical
access control. The iris is a thin circular diaphragm,
which lies between the cornea and the lens of the
human eye. The pattern of the human iris differs
from person to person, even between monocular
twins. The iris is considered one of the most reliable
and stable biometric, as it is believed to not change
significantly during a person’s lifetime. Iris
recognition is the most accurate
identification biometric.
Initially Flom and Safir proposed the automated
iris recognition [1]. The Iris recognition consists of
Munaga V N K Prasad
Institute for Development and Research in
Banking Technology
mvnkprasad@idrbt.ac.in
personal
five major steps i.e., iris acquisition, localization,
normalization, feature extraction and matching. A
critical step in an iris recognition system is to
segment automatically and reliably the iris from the
captured iris images. In previous segmentation
methods, most of which
integrodifferential operator or Hough transform
method. Wildes [2] used filtering and voting
procedure to segment the iris, which is the voting
procedure is realized via Hough transform on
parametric definitions of the iris boundary contours,
including pupil, limbus and eyelids boundaries.
Daugman [3] proposed an integrodifferential
operator for localizing iris region. The Daugman
system fits the circular contours via gradient ascent
together with radial Gaussian. Kong and Zhang [11]
proposed an accurate iris segmentation method for
reflection and eyelash detection. An edge detector is
applied to detect separable eyelashes, and intensity
variances are used to recognize multiple eyelashes.
A threshold and statistical model is proposed to
recognize the strong
respectively. Ma et al. [4] proposed an iris
segmentation method by edge detection and Hough
transform. Eyelids and eyelashes noises were not
considered in their method. Huang et al. [6] proposes
a new noise-removing approach based on the fusion
of edge and region information. Daugman rubber
sheet model [3] is a non-concentric normalization
method. In this method circler iris is represented into
rectangular form. The Wildes [2] system employs an
image registration technique, which geometrically
warps a newly acquired image, into alignment with a
selected database image. Masek [5] propose a new
non-concentric normalization model.
Daugman [3] extracted local texture phase feature
of iris by using 2D Gabor filters in order to generate
a 256-byte iris code and compared iris code using
the Hamming distance via the XOR operator. Wildes
[2] matched images using Laplacian pyramid multi
resolution algorithms and a Fisher classifier. Boles et
al. [13] have calculated zero-crossing representation
of 1-D wavelet transform at various resolution levels
of a virtual circle on an iris image to characterize the
texture of the iris. Lim et al. [7] uses the 2-D Haar
are based on
and weak reflection
Copyright © SITIS
- 831 -

Page 2

wavelet transform to extract the iris features into an
87-bit code and for classification they used a
modified competitive learning neural network (LVQ)
and Jafar et al. [9] also uses the 2-D Haar wavelet
transform to extract the iris features into an 87-bit
code. The disadvantage of the Gabor filter is that the
even symmetric filter will have a DC component
whenever the bandwidth is larger than one octave.
So Masek [5] and C.S Chin et al. [10] use the 1D
Log Gabor wavelets to encode the features because
the no DC components will pass the filters. This
paper deals with iris localization, normalization,
feature extraction and matching.
This paper is organized as follows. Section 2
provides an overview of our iris localization method
which identifies the inner and outer boundaries of
the iris. Section 3 explains how to transforming the
segmented iris region into fixed dimension
rectangular form iris. Section 4 presents noise
removing and enhancement. Section 5 and section 6
are present feature extraction which is done by using
2D Log-Gabor filters and matching which is done by
using simple binary XOR operation respectively.
Experimental results are reported in section 7 and
section 8 concludes this paper.

2. Iris Localization

The iris is a thin circular diaphragm, which lies
between the cornea and the lens of the human eye.
The acquired iris image has to be preprocessed to
detect the iris, which is an annular portion between
the pupil (inner boundary) and the sclera (outer
boundary). The inner and outer boundaries are non-
concentric circles. In this process first roughly
determine the iris region in the original image, and
then compute the exact parameters of the two circles
in the determined region. The detailed step-by-step
process is as follows:

1. First the captured image must be converted to
grayscale format if it is not already in an
intensity image. Subtract each pixel's intensity
of image from 255. This operation gives the
complement of image.
2. Identify the holes in the complemented image. A
hole is an area of dark pixels surrounded by
lighter pixels in complement image. Holes
identification is easy in complemented image
when compare to intensity image. If the holes
exist in the intensity image then these are reduce
the number of black pixels in the pupil. So fill
the holes in the
complemented image.
3. Apply the Canny edge detection operator [12] to
create the edge map image on the complemented
image. Canny edge detection is an efficient and
an easy method. Edge map image’s edges for
intensity image and
complemented image are very clear when
compare to intensity image’s edge map image
edges. Compute the average of all edge points.
This average edge point (xa, ya) as taken as a
rough center for the iris.
4. We binarize a 150x150 region of intensity
image centered at the point (xa , ya) by
adaptively selecting a reasonable threshold
using the gray level histogram of this
region[17]. Since the pupil is generally darker
than its surroundings, so in the binarized region
the pupil is in dark. Identify the number of
black pixels vertically and horizontally at each
coordinate in this region. Take the coarse center
of pupil (xc, yc) here xc and yc are the positions
of X-coordinate and Y-coordinate respectively
where the maximum number of black pixels
occur in this region.
5. By using this coarse center (xc, yc) apply the
Circular Hough Transform which is employed
by Wildes [2] in certain region on the edge
image to estimate the exact parameters of inner
and outer circles of iris. These parameters are
for pupil
00
(,)
pp
xy
is the center of pupil, rp is
the radius of the pupil circle, for outer circle
(,)
ii
xy
is the center and ri is the radius.
00
(a) (b)


(c) (d)


(d)
Figure 1. (a) Complemented iris image. (b) After holes
filled iris complemented image (c) Edge iris image. (d)
150 x 150 binarized image. (e) Inner boundary and
outer boundary detected iris image.

The results of the segmentation algorithm are
given in fig 1. Compared with the localization
method by Wildes [2] where the combination of
edge detection and Hough transform is also adopted,
- 832 -

Page 3

our method approximates the iris region before
Hough transform. This will reduce the search space
of Hough transform, and thus result in lower
computational cost. If the image contains dark places
at corners of image then first step could not give
rough center of pupil in Li Ma et al [4]. This
disadvantage can be overcome in our method.
Compared with the localization method by Li Ma et
al [4] our method is efficient method.

3. Normalization

Once the iris region is successfully segmented
from an eye image, the next stage is to transform the
iris region so that it has fixed dimensions in order to
allow comparisons. Normalization is a process of
transforming the segmented iris region into fixed
dimension. The dimensional inconsistencies between
eye images are mainly due to the stretching of the
iris caused by pupil dilation from varying levels of
illumination. Other sources of inconsistency include,
varying imaging distance, rotation of the camera,
head tilt, and rotation of the eye within the eye
socket. All these may affect subsequent feature
extraction and pattern matching. It is necessary to
compensate for these
normalization process will produce iris regions,
which have the same constant dimensions, so that
two photographs of the same iris under different
conditions will have characteristic features at the
same spatial location. In this paper, the iris ring is
map to a rectangular block of texture of a fixed size
(64x512). The circular iris inner and outer
boundaries are looks like in fig 2 two circles. In fig 2
angularly divided into 512 stripes and at every angle
strip radial divided into equal 64 stripes. At each
angle strip 64 iris pixel intensity values are
represented in vertically and angular stripes are
represented in horizontally as shown in fig 3. The
following formulas perform the transformation. Here
we consider the pupil center as the reference point as
shown in Fig 2.
xxr
=+
=+
θ=+
deformations. The
0
cos
ppp
θ
θ

(1)
0
sin
−
ppp
yyr

(2)
0
−
0
tan( )tan( )
ii
+
pp
yxyx
θ

(3)
(4)

00
222
()()
iiiii
xxyyr
−=

Where
00
(,)
pp
x
)
y
is the center of pupil circle,
00
(
radius of pupil circle and ri is the radius of limbic
circle. By using the equations (3) and (4) we can
easily get the (x i , yi ) values. This point (xi , yi ) is
the point on the iris circle at corresponding angle θ .
Now find the distance between point on the pupil
,
ii
xy
is the center of limbic circle, rp is the
circle (xp, yp) and point on the iris circle (xi , yi) by
using equation (5). Then divide this distance in equal
n stripes.
22
()()
ipip
Lxxyy
=−+−


(5
)

Now the rip = rp + L / n where n is the number of
stripes and we taken as 64. Where rip is the distance
between
00
(,)
pp
xy
From the
00
(,)
pp
xy
point on the iris at angle strip θ and radial strip L/n
which is rip distance away from
strip in radial the intensity values of iris are
representing in vertical. This process is continuing
for every angle strip and represent angle strip in
horizontally. So the normalized iris image is I(rip ,
θ ) as shown in fig 3 which represent the circler iris
as shown in fig 1.e into rectangular iris.

and first strip point at angleθ .
and rip we can easily identify the
00
(,)
pp
xy
. At each


Figure 2. Normalization process.


Figure 3. Segmented Iris Image with 64 x 512 size.

4. Iris image enhancement and denoising

The normalized iris image has low contrast and
may have non-uniform brightness caused by the
position of light sources. All these may affect the
subsequent processing in feature extraction and
matching. In order to obtain a more well-distributed
texture image, we first approximate intensity
variations across the whole image. The mean of each
16x16 small block constitutes a coarse estimate of
the background illumination. This estimate is further
expanded to the same size as the normalized image
by bicubic interpolation. The estimated background
illumination is subtracted from the normalized image
- 833 -

Page 4

to compensate for a variety of lighting conditions.
Then we remove high frequency noises by filtering
the image with a low-pass Gaussian filter. By this we
can see that finer texture characteristics of the iris in
Fig 4 become clearer than those in Fig 3. Fig 4
shows the preprocessing image.


Fig 4. After removing noise and enhancing
normalized iris image

5. Feature Extraction

Feature encoding is an important process in iris
recognition. In order to provide accurate recognition
of individuals, the most discriminating information
present in an iris pattern must be extracted. Only the
significant features of the iris must be encoded so
that comparisons between templates can be made.
Among all the iris recognition methods, Daugman’s
Gabor wavelet approach
recognizing iris patterns have been most widely used
in commercialized iris
Daugman extracted the local texture phase feature as
the iris code using complex 2D Gabor filters [3]. The
iris code size is 256 bytes long. Field [14] reported
that the Gabor filter has the disadvantage of DC
component whenever the bandwidth is larger than
one octave. Because the Gabor filters are not band
pass filters. One of the advantages of Log-Gabor
filters is that they are strictly band pass filters. So no
DC components will pass the filters.

In this paper convolving the normalized iris
pattern with 2D Log-Gabor filters generates iris
code. 2D Log-Gabor filters are in Cartesian
coordinate system of frequency domain [15] is given
in equation (6).

[lo g (
(,)e x p {
2[lo g (k

where
1
u
=
sin( )cos( ) vuv
θ= −+
orientation of the 2D Log-Gabor filter, u0 is the
center frequency, k determines the bandwidth of the
filter in the u1 direction , and σv determines the
bandwidth of the filter in the direction v1. This filter
function has no spatial domain expression. Here we
can construct the spatial domain filters by evaluate
the numerical inverse Fourier transform. After
performing inverse Fourier transform we can get the
real gr(x,y) and imaginary gi(x,y) responses of filter.
During the feature extraction the normalized iris
image is dividing into m by n blocks. Consequently,
the local phage information in each block is encoded
for encoding and
recognition systems.
2
2
10
1
22
v
0
/ )]
)]
} ex p {}
/2
uu
u
v
G u v
σ
−
−
=
(6
)
cos( )
θ
,
sin( )
is
uv
θθ+
,
1
θ the
into 2 bit codes according to the following equations
(7).
1 ( , )*hifI x yg x y dxdy
=
∫∫
Re
( , )0
r
xy
∫∫
≥

Re
0( , )*I x y( , )g x y dxdy 0
r
xy
∫∫
hif
=<

Im
1( , )*I x y ( , )g x y dxdy0
i
xy
∫∫
hif
=≥

Im
0( , )*I x y( , )g x y dxdy0
i
xy
hif
=<


(7)

Where * denotes the convolution operator, I(x, y)
denotes the normalized iris image, gr(x, y) is the
real part of the filter response and gi(x, y) is the
imaginary part of the filter response. The phase
information in each block is represented by 2-bit
codes and totally 2mn bits are generated for the
whole iris. In this paper we divide the image into 8 x
8 blocks. So the total number of bits in iris code is
1024.

6. Matching

Matching algorithm determines the similarity
between two given data sets. Applying the matching
algorithm on the input iris image and iris code
existing in the database does the iris verification.
The iris image is said to be authentic if the result
obtained after matching is more than the present
threshold value. In our approach we employed
hamming distance (8) to the similarity between two
iris codes.
⊗
∑
I
Where ⊗ denotes the Boolean Exclusive-OR
operator(XOR), maskA and maskB denote two iris
matching masks, respectively, “0” for the non-iris
regions, and “1” for the iris regions,I denotes the
AND operator. The predefined threshold makes the
decision whether two irises are from the same
person.

7. Experimental Results

In this paper we are experimented our approach
on the open iris database CASIA-IrisV3 [16].
CASIA-IrisV3 contains a total of 22,051 iris images
from more than 700 subjects. In our experiment we
consider 2000 iris images of 50 subjects in CASIA-
IrisV3-Lamp. The images are segmented using our
method. Segmentation is the main role in the iris
recognition. We segment the iris and normalized
with 64 x 512 size. By using our segmentation
method we extract the iris correctly from 1900
images out of 2000 images database i.e 95%. By
[()(
maskB
)]
()
codeAcodeB
maskA
maskAmaskB
HD
=∑
II

(8)
- 834 -

Page 5

using Li Ma et al [4] method we extract only 1670
images out of 2000 images database i.e 83.5%
because of the corners of the iris image is as dark as
with pupil. In the feature extraction 2D Log-Gabor
filter is used. Genuine Acceptance Rate (GAR) and
False Acceptance Rate (FAR) rates are calculated to
evaluate the performance of the system. The
database is tested for different threshold values to
calculate the GAR. A person is said to be genuine if
at least one iris matching score value is above the
threshold. FAR is calculated by matching each
image of a person with all the images of the
remaining persons in the database and the results are
given for different threshold values. Table 1 presents
the GAR and FAR at different threshold values for
2D Log-Gabor filters.

Table 1. GAR and FAR rates for different threshold
values for 2D Log-Gabor filter.

Threshold GAR
0.80 100%
0.84 99 %
0.87 98.36%
0.90 97%
0.92 95.84%
FAR
9.68%
5.05 %
2.94%
1.84%
0.63%
0.93 94% 0.31%

8. Conclusion

The objective of this work is to segment the iris
and extract the reliable features. A novel
segmentation method has been proposed which is an
enhancement over already existing segmentation
method [4]. By using our segmentation method, we
could extract the iris 95% images of the database
whereas using Li Ma et al [4] method we could
extract iris only 83.5% of the images from the same
database. A new normalization method has also been
proposed.

References

[1] L. Flom and A. Safir, “Iris Recognition system”, U.S.
Patent No.4641394, 1987.
[2] R. Wildes, “Iris Recognition: An emerging biometric
technology” Proceedings of the IEEE, vol.85, no.9,
September 1997.
[3] J. Daugman, “How
Proceedings of International Conference on Image
Processing, Vol. 1, 2002.
[4] Li Ma, Tieniu Tan, Yunhong Wang, Dexin Zhang,
“Efficient Iris Recognition by Characterizing Key
Local Variations”, IEEE Transactions on Image
processing, Vol. 13, June 2004, pp. 739-750.
iris recognition works”,
[5] Libor Masek, “Recognition of Human Iris Patterns for
Biometric Identification”, The University of Western
Australia, 2003.
[6] J. Huang, Y. Wang, T. Tan, et al., “A new iris
segmentation method for recognition”, Proceedings of
the 17th International Conference
Recognition, vol. 3, 2004, pp. 554–557.
[7] Shinyoung Lim, Kwanyong Lee, Okhwan Byeon, and
Taiyun Kim, “Efficient Iris Recognition through
Improvement of Feature Vector and Classifier”, ETRI
Journal, Volume 23, Number 2, June 2001.
[8] LiYua, David Zhangb and Kuanquan Wanga, “The
relative distance of key point based iris recognition”,
Pattern Recognition 40, 2007, pp. 423 – 430.
[9] Jafar M.H. Ali and Aboul Ella Hassanien, “An Iris
Recognition System to enhance E-security environment
Based on Wavelet Theory”, AMO-Advanced Modeling
and Optimization, Volume 5, Number 2, 2003.
[10] Chong Siew Chin, Andrew Teoh Beng Jin and David
Ngo Chek Ling, “High security Iris verification system
based on random secret integration”, Computer Vision
and Image Understanding 102, 2006, pp.169-177.
[11] W.K. Kong and D. Zhang, “Detecting eyelash and
reflection for accurate iris segmentation”, Int. J. Pattern
Recognition Artif. Intell. 17 (6), 2003, pp. 1025–1034.
[12] J. Canny, “A computational approach to edge
detection”, IEEE Trans. Pattern Anal. Mach. Intell. 8
(6), 1986, pp. 679–698.
[13] W.W. Boles and B. Boashah, “A Human Identification
Technique Using Images of the Iris and Wavelet
Transform”, IEEE Transaction on Signal Processing
Vol.46, 1998, pp.1185-1188.
[14] D. Field, “Relations between the statistics of natural
images and the response properties of cortical cells”, J.
Opt. Soc. Am.A/Vol: 4, 1987, pp. 2379 – 2394.
[15] Peng Yao et al, “Iris Recognition Algorithm using
modified Log Gabor Filters”, The 18th International
Conference on Pattern Recognition(ICPR’06), IEEE
Computer Society, 2006, pp. 461-464.
[16] URL: http://www.cbsr.ia.ac.cn/IrisDatabase.htm,
N. L. of Pattern Recognition (NLPR) Institute of
Automation (IA) Chinese Academy of Sciences
(CAS). CASIA iris image database (ver 3.0).
[17] Otsu N, “A threshold selection method from gray
level histograms”, IEEE Trans System, Man and
Cybernetics, SMC-9, 1979, pp. 62-66.
on Pattern

- 835 -