Efficient iris segmentation and pupil detection for visagisme applications

This paper explores the problem of automated iris segmentation and pupil detection. The motivation to develop such a system is that an accurate detection of the eye, iris and pupil provides useful information for many applications from various domains, such as medicine, biometrics, and fashion industry. The paper features the following contributions: training of a fast eye landmarks detector, an original and precise algorithm for an estimating the center of the pupil and the iris margin, especially where the eyelashes cover it, and the manual annotation of a new benchmark dataset for testing purpose. The proposed method is proved to be very accurate: for all detected points, the error is smaller than an approximation of the diameter of the pupil.