Locating the iris : A first step to registration and identification

Abstract

This paper presents a method for locating the iris borders in images of the cornea. It uses a combination of a priori information, statistics and active contours to both find the iris borders and assess its own success or failure. When applied to 107 images from 11 different people with varying eye colours including light blue and dark brown the algorithm reports 7 images for which it failed to find the borders, one of which was a false negative. In the other 100 images, the algorithm located the iris borders with average estimated errors of 5.0 pixels for the left iris-sclera border, 3.4 pixels for the right iris-sclera border and 2.1 pixels for the pupil-iris border. There were no false positive results.