REDUCING THE EFFECT OF PARTIAL OCCLUSIONS
ON IRIS RECOGNITION
Meryem Erbilek and Önsen Toygar
Computer Engineering Department, Eastern Mediterranean University
Gazimağusa, Mersin 10, Turkey
Keywords: Biometrics, iris recognition, feature extraction, multiple classifier systems, Principal Component Analysis.
Abstract: The difficulty in the process of human identification by iris recognition is that the iris images captured may
have occlusions by the eyelids and eyelashes. In that case, recognition of occluded iris patterns becomes
hard and the corresponding person may not be correctly recognized. In order to reduce the effect of eyelid or
eyelash occlusion on the recognition of human beings by their iris patterns, we propose a simple and
efficient method for iris recognition using specific regions on the iris images without using the traditional
preprocessing approach before applying the feature extraction method to recognize the irises. First of all,
these regions are individually experimented and then the outputs of each region are combined using a
multiple classifier combination method with the feature extraction method Principal Component Analysis
(PCA). The experiments on the iris images, with and without occlusions, demonstrate that the proposed
approach achieves better recognition rates compared to the recognition rates of the holistic approaches.
1 INTRODUCTION
Reliable person identification or verification is
becoming a very important topic in recent years in
most of the countries in the world. Biometrics is a
popular field of study which is related to the
recognition of individuals based on their
physiological and/or behavioral characteristics. Iris
recognition is one of the most secure and reliable
biometric recognition systems (Ma et al., 2004; Cui
et al., 2004; Daugman, 2006) to recognize human
beings. It is not suitable to get iris images of people
in each condition and enviroment to recognize
human since the iris images must be captured by a
special camera from people who are willing to do
this. However, for controlled enviroments such as
airports, iris recognition seems to be a convenient
biometrics to identify human beings.
The difficulty in the process of human
identification by iris recognition is that the iris
images captured may have occlusions by the eyelids
and eyelashes. In this respect, recognition of
occluded iris patterns becomes hard and the
corresponding person may not be correctly
recognized.
In order to reduce the effect of eyelid or eyelash
occlusion on the recognition of human beings by
their iris patterns, we propose a simple and efficient
method for iris recognition using specific regions on
the iris images without using the traditional
preprocessing approach before applying the feature
extraction methods to recognize the irises. In our
method, the iris image is partitioned into vertical left
and right regions and also top and bottom regions of
the iris are partitioned horizontally into two regions.
In this way, the horizontal or vertical regions of the
iris pattern without including the pupil is obtained
from the iris image. Firstly, these regions are
individually experimented on the CASIA iris dataset
and the recognition rate of each region is recorded
using Principal Component Analysis (PCA) as the
feature extraction method. Then, an appropriate
multiple classifier combination method is used to
combine the output of these regions. Various
experiments are performed to test the recognition
performance of the classical and the proposed
approaches using iris images with and without
occlusions on different regions of the irises. The
results are presented in the further sections.
The rest of the paper is organized as follows.
Section 2 describes PCA which is the feature
extraction method used in this study. The proposed
approach for iris recognition based on the
partitioning method is described in Section 3. The
543
Erbilek M. and Toygar Ö. (2008).
REDUCING THE EFFECT OF PARTIAL OCCLUSIONS ON IRIS RECOGNITION.
In Proceedings of the Third International Conference on Computer Vision Theory and Applications, pages 543-546
DOI: 10.5220/0001084305430546
Copyright
c
SciTePress
experimental study and the discussions are presented
in Section 4. Finally, Section 5 concludes the paper.
2 FEATURE EXTRACTION
METHOD
Appropriate feature extraction is an essential
component of a successful biometrics recognition
algorithm. For this purpose, statistical
dimensionality reduction methods such as Principal
Component Analysis, Linear Discriminant Analysis
and Independent Component Analysis are
demonstrated to be successful in several academic
studies and commercial applications (Toygar and
Acan, 2004, Chu and Chen, 2005). The success and
popularity of these algorithms are mainly due to
their statistics–based ability of automatically
deriving the features instead of relying on humans
for their definitions. These algorithms are widely
studied for the recognition of human beings using
their physiological characteristics such as the face,
iris, fingerprint and so on.
In this study, we apply PCA which is one of the
most popular methods used in the literature for
feature extraction on different regions of the iris
images. PCA projects images into a subspace such
that the first orthogonal dimension of this subspace
captures the greatest amount of variance among the
images and the last dimension of this subspace
captures the least amount of variance among the
images (Kirby and Sirovich, 1990; Turk and
Pentland, 1991). In this respect, the eigenvectors of
the covariance matrix are computed which
correspond to the directions of the principal
components of the original data and their statistical
significance is given by their corresponding
eigenvalues.
3 PARTITIONING APPROACH
ON IRIS RECOGNITION
Partitioning approach for the iris recognition
problem is applied by dividing the iris images into
four segments and the features of each segment are
extracted independent of each other using PCA
method.
All the training and testing face images used in
the experimental study are cropped as shown in
Fig.1. Cropping operation is applied in the same way
for both the training and test images, so that all the
images include only the iris and the pupil after this
operation. Partitioning approach is applied after this
operation and divides the image into four segments,
each segment including the horizontal top,
horizontal bottom, vertical left and vertical right
regions of the iris excluding the pupil from the iris
as shown in Fig.1. In this way, the iris patterns of the
training and test images are stored in four different
partitions or segments.
Figure 1: Partitions of an iris image: left, right, top and
bottom regions.
After applying PCA algorithm on each segment
of training and test images, Euclidean distance
measure is used to find the distance between features
of these image segments. For each test image, the
distances between the test image and all the training
images are compared. The training image that has
the minimum distance to the test image is the image
that mostly resembles to it.
In this study, PCA is implemented with a divide-
and-conquer strategy for the solution of the iris
recognition problem. The divide-and-conquer
approach implemented over multiple classifier
systems (MCSs) (Ho et al.,1994; Kittler et al.,1998)
is used to improve the computational efficiency and
recognition performance of PCA method on the iris
recognition problem. MCSs combine the output
information provided by two or more classifiers.
In the implementation of the divide-and-conquer
methodology for the iris recognition problem, iris
images are divided into two horizontal (top and
bottom regions) and two vertical (left and right
regions) segments and PCA is applied on each iris
segment as a feature extraction method.
Consequently, a multiple classifier system
is established based on a particular distance
measure. Finally the outputs of multiple classifiers
are combined using a well-known multiple classifier
combination method to recognize the whole iris
under occlusions.
4 EXPERIMENTAL RESULTS
The experiments were done on the iris images
chosen from the third version of CASIA iris dataset
VISAPP 2008 - International Conference on Computer Vision Theory and Applications
544
(CASIA-IrisV3). The performance of the proposed
approaches are tested using PCA algorithm applied
on the individual regions of the iris images. PCA
approach is applied in the same way as explained in
(Toygar and Acan, 2004). In this study, the iris
images used are cropped so that they only include
the iris and pupil regions. The iris images were
scaled down to 199x156 pixels from the original size
of 640x480 pixels. The experiments were done using
100 training and 100 test images (two samples per
person).
In the implementation of the proposed
approach, multiple classifiers are considered using
the Majority Voting classifier combination method.
The output of each individual classifier is computed
separately, followed by the multiple-classifier
combination procedure which produces the final
classifier or recognition output. Majority Voting
method selects the class which receives the largest
number of votes as the majority decision (Kittler et
al.,1998). In case of equality of the votes, one of the
classifiers with maximum number of votes is
selected arbitrarily.
Four set of experiments are performed and the
results are presented in Table 1 through Table 4. The
recognition rates of the holistic PCA (PCA applied
on the whole iris image), partitioned PCA (PCA
applied on the four regions of the iris image
individually) and the proposed approach
(combination of the four regions using a MCS) are
recorded. The results are presented using the images
without occlusions and the images with occlusions
on the training and test images.
In the first set of experiments, the whole iris
images are experimented using the holistic PCA
method, partitioned PCA on the left, right, top and
bottom regions individually and the proposed
approach to combine the outputs of each individual
region of the iris image.
As shown in Table 1, the recognition rate using
the holistic PCA method is 84% and the three
partitioned PCA methods achieves less than this
recognition rate while partitioned PCA applied on
the bottom region of iris achieves 88% which is
better than the holistic counterpart. Additionally, the
proposed approach combines outputs of the
individual regions and the recognition rate is
Table 1: Recognition rates on iris images without
occlusion.
Approach Recognition Rate (%)
Holistic PCA (Whole iris) 84
Partitioned PCA (Left) 80
Partitioned PCA (Right) 66
Partitioned PCA (Top) 78
Partitioned PCA (Bottom) 88
Proposed Approach 96
Table 2: Recognition rates on iris images with occlusions
on top of the images.
Recognition Rate (%)
using occlusion of
Approach
3% 5% 7% 10%
Holistic PCA (Whole iris) 70 80 70 66
Partitioned PCA (Left) 76 70 70 70
Partitioned PCA (Right) 76 78 82 84
Partitioned PCA (Top) 56 46 50 44
Partitioned PCA (Bottom) 54 40 40 32
Proposed Approach 80 82 76 70
Table 3: Recognition rates on iris images with occlusions
on bottom of the images.
Recognition Rate (%)
using occlusion of
Approach
3% 5% 7% 10%
Holistic PCA (Whole iris) 82 80 72 60
Partitioned PCA (Left) 80 80 66 66
Partitioned PCA (Right) 78 80 84 72
Partitioned PCA (Top) 68 62 44 38
Partitioned PCA (Bottom) 64 48 48 48
Proposed Approach 92 84 80 68
Table 4: Recognition rates on iris images with occlusions
on top and bottom of the images.
Recognition Rate (%)
using occlusion of
Approach
3% 5% 7% 10%
Holistic PCA (Whole iris) 74 72 62 52
Partitioned PCA (Left) 70 68 60 62
Partitioned PCA (Right) 80 76 78 68
Partitioned PCA (Top) 40 36 32 14
Partitioned PCA (Bottom) 36 42 24 22
Proposed Approach 74 70 68 64
increased to 96% which is a significant
improvement.
On the other hand, the iris images are occluded
artificially with different percentages (3%, 5%, 7%,
10%) on different regions of the iris such as the top,
bottom and both of them. As an example, 10%
occlusions on top, bottom, and top and bottom
regions are shown in Figure 2.
REDUCING THE EFFECT OF PARTIAL OCCLUSIONS ON IRIS RECOGNITION
545
Figure 2: Iris images (a) without occlusions and with
occlusions on (b) top (c) bottom and (d) top and bottom
regions.
In the second set of experiments, the iris images
are occluded from the top region of the iris. The
recognition rates for various occlusion percentages
are presented in Table 2. The data on this table
shows that some of the partitioned PCA approaches
slightly increase the performance of the holistic
PCA. The recognition rate of the proposed approach
is significantly increased compared to holistic
counterparts.
The third and the fourth set of experiments also
demonstrate that the proposed approach by
combining the outputs of individual regions achieves
better performance compared to the performance of
the holistic PCA and partitioned PCA methods.
In general, the experiments using holistic PCA,
partitioned PCA and the proposed approach without
occlusions demonstrate that the proposed approach
achieves a significant improvement for the
recognition of iris images (from 84% to 96%). On
the other hand, for the iris images with occlusions on
either top or bottom of the iris, the recognition rates
demonstrate that the proposed approach is not
sensitive up to 5% occlusions. The results obtained
by the proposed approach with 5% occlusions on
either top or bottom of the irises are equivalent to
the results obtained by holistic PCA without
occlusions.
5 CONCLUSIONS
A simple and efficient approach for reducing the
effect of partial occlusions and improving the
recognition performance of the iris images is
presented. The experiments performed using holistic
PCA, partitioned PCA and the proposed approach
without occlusions demonstrate that the proposed
approach achieves a significant improvement for the
recognition of iris images. It can also be stated that
the proposed approach is not sensitive up to 5%
occlusions on top or bottom regions of the iris
images compared to the holistic PCA method
without occlusions.
ACKNOWLEDGEMENTS
Portions of the research in this paper use CASIA
IrisV3 collected by the Chinese Academy of
Sciences’ Institute of Automation (CASIA).
REFERENCES
CASIA-IrisV3, http://www.cbsr.ia.ac.cn/IrisDatabase.htm
Chu, C.T., Chen, C.H., 2005. High Performance Iris
Recognition Based on LDA and LPCC. In ICTAI’05,
Proceedings of the 17
th
IEEE International
Conference on Tools with Artificial Intelligence, Hong
Kong.
Cui, J., Huang, J., Tan, T., Sun, Z., 2004. An iris image
synthesis method based on PCA and super-resolution.
In ICPR’04, 17th International Conference on Pattern
Recognition, Cambridge, UK.
Daugman, J., 2006. Probing The Uniqueness and
Randomness of IrisCodes: Results from 200 Billion
Iris Pair Comparisons. Proceedings of The IEEE,
Vol.94, No.11, pp.1927-1935.
Ho, T., Hull, J., Srihari, S., 1994. Decision combination in
multiple classifier systems, IEEE Transactions on
Pattern Analysis and Machine Intelligence, Vol.16,
pp. 66-75.
Kirby, M. and Sirovich, L., 1990. Application of the
Karhunen-Loeve Procedure for the Characterization of
Human Faces. In IEEE Trans. Pattern Analysis and
Machine Intelligence. Vol.12, pp. 103-107.
Kittler, J., Hatef, M., Duin, R.P.W., Matas, J., 1998. On
Combining Classifiers, IEEE Trans. Pattern Analysis
and Machine Intelligence, Vol.20, No.3, pp.226-239.
Ma, L., Tan T., Wang Y., Zhang D., 2004. Efficient iris
recognition by characterizing key local variations. In
IEEE Transactions on Image Processing. Vol.13,
no.6, pp.739-750.
Turk, M. and Pentland, A., 1991. Eigenfaces for
Recognition. In Journal of Cognitive Neuroscience.
Vol.3, pp.71-86.
Toygar, Ö. and Acan, A., 2004. Multiple Classifier
Implementation of a Divide-and-Conquer Approach
Using Appearance-Based Statistical Methods for Face
Recognition. In Pattern Recognition Letters, Vol.25,
pp. 1421-1430.
VISAPP 2008 - International Conference on Computer Vision Theory and Applications
546