Research On Iris Recognition Algorithms Under Non-ideal Situation

The identity safety is an important topic in modern society． Traditionalrecognition means cannot keep up with the pace of modern society. The emergingbiometric recognition technologies are more and more popular by their own uniqueadvantages. Among all kinds of biometric recognition technologies, uniqueness andstability of iris recognition are both excellent, and iris is not easy to be changed, easyto do living inspection. So iris is quite suitable for identification. An iris recognitionsystem should have both a relatively high recognition rate and processing speed.However, the current researches of iris recognition are not enough to meet the tworequirements at the same time. Therefore, we do some research on iris recognitiontechnology to improve the performance of our iris recognition system.This paper mainly studies biometric recognition algorithms based on the non-ideal iris images. To the iris images with a lot of speckles, this paper proposes an algorithm to detect speckles in pupil area based on2D Gabor filters first. Then the speckles are replaced selectively with the average intensity of a fixed square region. Finally, a novel evaluation index is introduced, which improves the pupil location accuracy rate ofour former pupil location algorithm. In the widely used CASIA v3.0iris database, thismethod has achieved a good result.To detect eyelid and eyelash, we select a fixed detection model. Then we move the model respectively in the regions of the upper and lower eyelids to achieve the bestcover effect. So we can detect automatically the upper and lower eyelids.We use SIFT algorithm to extract the iris features which is not sensitive to the accuracy of iris location. SIFT describes the local characteristics of images, it has invariance to the rotation, scale and illumination changes. At the same time, it has certain stability to the noise, perspective and affine transformation. SIFT is one of the best local feature descriptors.In this paper, AdaBoost is introduced into the iris location. After eyes detection,we screen eyes by size and histogram methods, at last locate the iris images. This paper detects eyes based on AdaBoost for narrowing the search scope and improving the speed and accuracy of iris location. In CASIA v4.0iris database, we get a good effect on iris location.