Lef3a: Pupil segmentation using Viterbi search algorithm

A non-circular pupil segmentation approach using Viterbi path research is presented. Around the circle approximation of the pupil, a large region of gradient is produced radially. An enhancement step is used in order to remove outliers from the gradient image. A Viterbi algorithm is applied to find the contour that maximizes the gradient value along a connex contour. Our method which we called Lef3a (a snake that lives in Tunisian desert) is compared to three state of the art methods: a circular detector, an elliptical detector and a Snake detector. The two first methods are developed by our team; the last method is an external work based on Daugman's published method. We will show some segmentation results on non circular pupils, and we will compare the performance of all segmentation methods coupled with a standard iris coding and matching process on Notre Dame Database and our private database. Finally all methods are compared in terms of time processing.