A computational efficient iris extraction approach in unconstrained environments

Abstract

This research introduces a noise-resistant and computational efficient segmentation approach towards less constrained iris recognition. The UBIRIS.v2 database which contains close-up eye images taken under visible light is used to test the proposed algorithm. The proposed segmentation approach is based on a modified and fast Hough transform augmented with a newly developed strategy to define iris boundaries with multi-arcs and multi-lines. This optimized iris segmentation approach achieves excellent results in both accuracy (2% error) and execution speed (≤0.5s / image) using a 2.4GHz Intel® Q6600 processor with 2GB of RAM. This 2% error is an Exclusive-OR function in term of disagreeing pixels between the correct iris considered by the NICE.I committee and the segmented results from the proposed approach. The segmentation performance was independently evaluated in the "Noisy Iris Challenge Evaluation", involving 97 participants worldwide, and ranking this research group in the top 6.