An improvement of iris-blob map approach for iris identification

This paper proposes an improved algorithm to identify iris with an iris-blob map. The localized iris segment is unwrapped and transformed from a Cartesian-coordinate image to a polar-coordinate image. In this work, iris texture is enhanced with the Difference of Gaussian (DoG). After thresholding the DoG images, texture blobs (or regions) are located and bounded by rectangles. Then, the occlusions from eyelid and eyelashes are filtered out. Our iris feature is a map (or template) consisting of iris texture bounding rectangles and Hu moments of the detected blobs. With the fusion of our new feature and new matching score, EER is greatly reduced from 19.5% to 4.8%.