Iris recognition with tunable filter bank based feature

In this paper, a tunable filter bank is proposed to extract region based features from non-cooperative iris images. The proposed method is based on half band polynomial of 14th order. The existing iris recognition algorithms work well with highly good quality images acquired under constraint and cooperative environments. These well-known techniques fail to perform well, when the raw images encounter segmentation failure during preprocessing. Their accuracy falls drastically, especially when the iris images are highly occluded due to different artifacts. Apart from that, segmentation failure during iris normalization makes the feature extraction more difficult. In our work, a tunable biorthogonal filter bank is proposed, for which the filter co-efficients are extracted from polynomial domain instead of z -domain. The proposed filter bank provides an opportunity to tuning and optimize the filter co-efficients. Experimental results using publicly available databases like CASIAv3, UBIRISv1, and IITD show the superiority of the proposed feature over the existing ones given its low template size, low feature extraction and feature matching time. Though accuracy yielded by the proposed filter bank is of the same order as found with existing features, achieving similar accuracy as state-of-the-art methods with less time and computation is a substantial development due to real-time usage of biometric systems.