A Pupil Detection Algorithm Based on Contour Fourier Descriptors Analysis

A high detection rate pupil detection algorithm is presented. This algorithm includes a preprocessing phase of the dark pupil image consisting of blurring, binarization and morphological operations. Next, during the processing phase, for each pupil candidate contour the Fourier descriptors of the convex-hull enclosure are determined. Each Fourier descriptor is used to obtain a metric using the Total Harmonic Distortion (THD) measure. This metric describes the similarity degree of the considered contour to an ideal ellipse. The algorithm selects the contours that fulfill the THD based metric condition and chooses the largest one as the pupil. Finally, to accurately estimate the pupil center, the Least Square Fit to Ellipse (LSFE) procedure is used. The proposed algorithm achieves over 85% average five pixels detection rate for three representative databases. The main advantage of the method is that the proposed THD based metric can be used for both circular and elliptical shaped pupil detection.