Automatic iris mask refinement for high performance iris recognition

How to estimate artifacts on an iris image in polar domain is an important question for any iris recognition system which pursues high recognition rate as its goal. In literature, there are many different existing algorithm that estimate iris occlusion in either Cartesian or polar coordinate. In this paper, our goal is not to propose another new method to compete with existing method. Rather, our goal is to propose a new algorithm which can take any iris mask estimated by existing algorithm, and refine it into a much more accurate mask. In this way, our proposed method could co-work with any other existing algorithm and improve iris recognition performance. Experimental results show our proposed method can improve iris recognition rate by a great lead compared to the performance of the system using the unrefined iris masks.