Fast and robust probabilistic inference of iris mask

Iris masks are essential in iris recognition. The purpose of having a good iris mask is to indicate which part of iris texture map is useful and which part is occluded or contains noisy artifacts such as eyelashes, eyelids and specular reflections. The accuracy of the iris mask is extremely important. The performance of the iris recognition system will decrease dramatically when iris mask is inaccurate, even when the best recognition algorithm is used. Traditionally, people used naive rule-based algorithms to estimate iris masks from the iris texture map. But the accuracy of the iris mask generated in this way is questionable. In this paper, we propose a probabilistic and learning-based method to automatically estimate iris mask from iris texture map. The features used in this method are very simple, yet the resulting estimated iris mask is significantly more accurate than the rule-based methods. We also demonstrate the effectiveness of the algorithm by performing iris recognition based on masks estimated by different algorithms. Experimental results show the masks estimated by the proposed algorithm help to increase the iris recognition rate on NIST Iris Challenge Evaluation (ICE) database.