Study On Noise Detection In Iris Image

Iris recognition has been becoming an active topic in biometrics due to its high reliability for personal identification. The occlusions of eyelid and eyelash in iris image are two kinds of noise to be difficult to detect and lower the performance of the system. In this dissertation, detections of these two kinds of noise are studied. Each crucial components of iris recognition system is described. Detailed analysis is given about advantages and disadvantages of the current occlusion detecting algorithms. Two gray-scale morphological noise detection algorithms are presented for the eyelid and eyelash occlusions, respectively. Firstly, an arc morphological structuring element is designed for detecting eyelid edge. A set of candidate points for eyelid edge can be obtained by gray-scale morphological opening, image segmenting, and edge detecting. Then the eyelid edge is fitted on the basis of Bezier curves. Secondly, a crossed morphological structuring element is developed. An iris image, whose intensity mostly distributes in several sections, can be acquired after gray-scale morphological opening. Thus a binary image of eyelashes is obtained. The experimental results indicate that the proposed algorithms have better effectiveness on detecting these two kinds of occlusion noises, and are helpful to decrease the Equal Error Rate of iris recognition system and improve the discriminability of iris patterns.In addition, an algorithm for iris recognition is also discussed. After analyzing the amplitude-frequency responses of real and imaginary part of the 2-D complex Gabor filter, we find that the 2-D odd Gabor filters is more effective than 2-D complex Gabor filter in extracting the features of iris texture. Then an algorithm for encoding phase information of iris texture with 2-D odd Gabor filters is introduced. Pattern matching is realized by computing Hamming distance between two templates. The experimental results show that the Equal Error Rate (EER) of system by using this method has about 12% decrease compared with the method for encoding phase information of iris texture based on 2-D complex Gabor filters.