Research On Partial Similarity Measure Iris Recognition Alogrithm Based On Local Features

With the increasing of requirement for information security, traditional identification technologies are insufficient for people'need in many applications. Nowadays, more and more biometrics character technologies have been applied in identification recognition. Iris recognition is a biometric technology, which has special biometric characteristic and is stable in recognition. Compared with other biometrics technologies, iris recognition has become more important recently.This paper do some research by analyzing and finishing the previous work on iris recognition, and propose some new ideas in image preprocessing and feature extraction.Firstly, an accurate and fast iris location algorithm based on bi-constraints of iris structure and edge distribution is proposed in the paper. For the iris inner edge location, the least square circle fitting method is used there. For the outer edge location, an adaptive template which constructed based on the pupil's position parameters and radius parameters that obtained from the first step is used to determin the effective area where the outer boundary existed. Finally the Hough transform is used to locate the outer boundary precisely.Secondly, to capture the multi-scale and multi-orientation features of iris images, an iris recognition algorithm combining steerable pyramid decomposition with centralized local binary pattern texture descriptor is proposed in this paper. Firstly, the proposed method decomposes the normalized iris images by convolving them with multi-scale and multi-orientation steerable pyramid filters to extract their corresponding filtered sub-images. Then, the centralized local binary pattern is used to extract the local neighbor pattern of each sub-image and form a histogram sequence which can be used to describe the iris texture feature. Experimental results demonstrate the effectiveness of the method.Lastly, to obtain better recognition rate in small training samples, an iris recognition method based on local centralized binary pattern and partial similarity measure is presented in this paper. The partial similarity measure used there can not only assure the recognition rate, but also reduce the complexity in iris recognition. This method is feasible in both theory and experiment, and gets a better recognition rate under one training sample.