Page 9 - tmp

Application of texture recognition using three-dimensional textons to iris biometric
authentication
Lazar Mitrović

Center for talented youth Belgrade II, Belgrade, lazanet96@gmail.com

1. Introduction

The iris is a muscular circular structure in the eye. It is
responsible for controlling the diameter and size of the
pupil – so as the amount of light reaching the retina. Color
of iris determined by more than one gene, but iris pattern is
developed in uterus in period from three to eight months
after conception. It is distinctive by each iris and represents
Figure 1 - Efficiency of different filter banks (training set - 15
different positions, numbers and sizes of iris arteries. Iris irises).
pattern is same regardless of age and it is impossible to
surgically modify it without risking loss of vision and as
number of possible pattern combinations is several times
greater than whole world population, all these things make
iris good candidate for biometric authentication.

2. Method

Most popular method for iris recognition is one suggested
by J.Daughman. In this research we study usability and Figure 2 - Efficiency of different filter banks (training set - 15
performance of Iris recognition technique based on texture irises).
recognition algorithm by Thomas Leung and Jitendra
Malik. Advantage of this method is that it uses entire filter
bank, so it produces more features for comparison. This is 4. Conclusion
the first attempt of this kind of application for that
algorithm. Implementation is done in MatLab environment
and combined with iris normalization by converting Implemented method is shown to be usable for iris
cartesian to polar coordinates, and produces fair results recognition with over 72% success rate. However, we also
comparing to conventional method. Also multiple different saw that this method is not as precise as Daugman’s
parameters for number of clusters in K – Means are tested because of false positives on badly defocused examples.
to select one that is most appropriate. This rate could be increased by using higher resolution
samples in controlled environment, different threshold for
histogram comparison and automatic segmentation
algorithm with eyelashes exclusion.

References

[1.] An Improved Approach for IRIS
Authentication System by Using
Daugman’s Rubber Sheet Model,
Image 1 – Segmented and normalised iris used in Segmentation, Normalization and RSA
implementation Security Algorithm, Mininath Raosaheb
Bendre, Sandip Ashok Shivarkar
3. Results [2.] How Iris Recognition Works, John

Daugman, PhD, OBE
Algorithm is trained on 20 classes (different irises) with 10- [3.] Representing and Recognizing the
20 iris pictures for each subject and testing was done on 65 Visual Appearance of Materials using
randomly selected irises. Also S, MR8 and LM filter banks Three-dimensional Textons, Thomas
were used. Collected results are shown on Figures 1 and 2. Leung, Jitendra Malik