An Iris Recognition System Based on Angular Radial Partitioning and Statistical Texture Analysis with Sum & Difference Histogram

Abstract

Iris based identification systems are considered among the most promising recognition systems due to the inner characteristics of the iris, such as uniqueness, stability and time invariance. This paper proposes a new texture based iris recognition system based on Angular Radial Partitioning (ARP) and Sum & Difference Histogram (SDH). After the iris segmentation step, ARP is used to divide the iris’s texture into sectors, SDH allows for the production of probability vectors, which are then used to extract statistical features. Finally, classification is performed with the K-Nearest Neighbour algorithm. Experimental results on the Ubiris and Upol databases testify the superior performance of the proposed approach, which can handle the presence of eyelids and eyelashes, as well as partially occluded irises and out of focus images. In all experiments the accuracy of the our system is around 97% also when the training set is made up of only two pictures per class, and the corresponding low percentage of FAR suggests that the proposed approach is a good prototype for biometric recognition systems run in identification mode.