Human identification using correlation metrics of iris images

This paper presents work done based on second order statistical features including cross- and auto-correlations as well as co-occurrence matrices of iris images in an attempt to extract a simple, yet powerful, set of features of an iris as a biometric. Prior to our work, the most prevalent methods for iris identification include the frontier work based on the use of quadrature 2-D Gabor wavelets with the Hamming Distance-based classification [1,2], circular Gabor filters with a nearest feature line (NFL) classifier [3], dyadic wavelet transform with the zero-cross detectors [4], texture analysis [9] and transient signal [11], and independent component analysis (ICA) [7], and boundary localization [10]. Our method differs significantly from the earlier approaches to iris recognition in that it relies on the wide-sense stationary approximation to the texture and gray-level characteristics of irises, and aims to lend itself for a single-chip hardware implementation. Our preliminary results show that cross- and auto-correlators along with co-occurrence matrix features are highly likely to be prominent iris discriminators.