Iris Identification Method Based On Scale Invariant Feature Transformation

Iris identification which is a part of biometric identification, it’s the uniqueness, universality, stability, collect-ability and acceptability of it makes its identification error rate is the lowest compared to all other biological recognition technologies. The rapid development of iris recognition in recent10years makes this technology achieved satisfying results in the field of core algorithm of the iris recognition system. But as the demands updating constantly, the iris recognition technology is also required to improve accordingly.Although the recognition rate of many iris recognition methods is considerably high, but they are applied in ideal conditions aimed at specific iris image which is collected by specific collection devices, and they are only applied in limited areas. Hence the main direction of scholars in this area is making iris recognition method to get stronger anti-interference ability. By studying the latest theory of scale invariant feature transformation and researching the expression of the samples belong to the same kind in multi-scale space, applies scale invariant feature transformation method into the iris recognition process, and expects to eliminate the influence of the iris image translation, rotation, illumination change and so on. By comparing with the classical iris recognition method through experiment, the goal of this paper is to work out the iris recognition method of strong stability in a practical environment. The major research contents and results in this paper are as follows:1. Researches the existing iris quality evaluation methods, selects subjects with coarse and fine evaluation method in preprocessing stage. Studies the existing iris image localization algorithm, applies the traditional localization algorithm and the level set algorithm into the iris image localization, and analyzes the different influences of it to the experimental process, and we found that the level set localization method has obvious advantages towards to solving the problems of eyelid shade, so this method is adopted in this paper. Researches the image enhancement algorithm, applies the part histogram and morphology enhancement method into the iris image, and compares the different enhancement effects.2. Studies the SIFT (scale invariant feature transformation) feature extraction algorithm, and applies it into the iris image recognition, and analyzes the characteristics of the iris feature extraction (strong stability of position, scale and direction. avoids the effects of image normalization to image feature recognition). Researches the effects of the important parameters in the SIFT algorithm to feature extraction, such as Gaussian scale factor, contrast threshold, and curvature threshold, and adjusts the parameters with the best experiment.3. Studies the SIFT feature matching method, changes the simple matching the number of key to comparing the proportion of key points when considers about the number of the key points of images of different human eye is different. Does dimension reduction experiments to SIFT feature descriptor. changes the Euclidean distance matching of SIFT algorithm to block matching for higher efficiency, and implemented above methods by matlab programming.4. Applies three iris image bases, which are CASIA-V1.0, CASIA-V3-Interval, MMU, as the experimental objects, and selects the template library which named as TEST with better matching effect to accurate experiment, the result shows5%increases in recognition rate accuracy.5. Studies the other classic iris recognition algorithms, such as wavelet zero crossing algorithm, two-dimensional wavelet multi-scale decomposition algorithm. Compares different methods through simple experiment and analyzes the algorithm.To sum up, this paper improved the SIFT algorithm and applied it into the iris feature extraction, and completed feature matching method. Moreover, experiments are conducted and the quality evaluation, image localization, image enhancement method in the iris preprocessing part are analyzed in this paper. And experiments results proves that this method is able to obtain satisfying recognition results.