Iris recognition based on a novel multiresolution analysis framework

In this paper, we proposed an efficient multiresolution decomposition framework for iris texture feature extraction, which has superior directionality and localized frequency partition. It consists of an improved circular symmetric filter bank followed by a directional filter bank. In our iris recognition system, we adopt a novel non-polar coordinate normalization strategy as iris preprocessing method. It preserves and enhances the geometric structure of an original iris image and is suitable for multiscale geometric analysis. We design the multiresolution analysis framework according to the texture distribution of normalized iris images, and partition them into multi-scale and multi-directional subbands. Then, we employ the normalized energy components as elements of the feature vector and use support vector machine (SVM) to classify the features. Experimental results show that the proposed method can achieve high classification accuracy with fewer features.