Low Cost Eyelid Occlusion Removal to Enhance Iris Recognition

Abstract

The Iris recognition system is claimed to perform with very high accuracy in given constrained acquisition scenarios. It is also observed that partial occlusion due to the presence of the eyelid can hinder the functioning of the system. State-of-the-art algorithms consider that the inner and outer iris boundaries are circular and thus these algorithms do not take into account the occlusion posed by the eyelids. In this paper, a novel low-cost approach for detecting and removing eyelids from annular iris is proposed. The proposed scheme employs edge detector to identify strong edges, and subsequently chooses only horizontal edges. 2-means clustering technique clusters the upper and lower eyelid edges through maximizing class separation. Once two classes of edges are formed, one indicating edges contributing to upper eyelid, another indicating lower eyelid, two quadratic curves are fitted on each set of edge points. The area above the quadratic curve indicating upper eyelid, and below as lower eyelid can be suppressed. Only non-occluded iris data can be fetched to the further biometric system. This proposed localization method is tested on publicly available BATH and CASIAv3 iris databases, and has been found to yield very low mislocalization rate.