Adaptive and localized iris weight map for accurate iris recognition under less constrained environments

Accurate iris recognition from the distantly acquired face or eye images requires development of effective strategies which can account for significant variations in the segmented iris image quality. Unlike conventional stop-and-stare mode iris recognition, iris images acquired under less constrained imaging environment specially those under visible illumination, are degraded by multiple sources of noise. Such accompanying noise is often embedded in the feature space (say iris code), during the feature extraction phase following the image normalization, and is highly correlated with the consistency of resulting iris bits. The bit consistency can be learned from the temporal iris codes such that the stable bits are considered less likely to be corrupted by noise, or vice versa. The relationship between the bit consistency and the accompanying noise is proposed to be modeled using a non-linear relationship that can follow a power-law. Therefore the noise perturbed bits are given less emphasis while consistent bits are given higher weight. The superiority of proposed iris matching strategy is ascertained by providing comparison with other state-of-the-art algorithms using publicly available databases: CASIA.v4-distance and UBIRIS.v2. Our experimental results suggest improvement of 13.9% and 47.2% in the average rank-one recognition respectively for the CASIA.v4-distance and UBIRIS.v2 databases.