Abstract
Videooculography is an eye tracking method widely used in vision researches. The essence of videooculography is to estimate pupil center accurately, irrespective of drooping eyelids, eyelashes, corneal reflection, and non-uniform lighting. Commonly, eye image is binarized and the contour of pupil is extracted. Ellipse detection finds the center with a high degree of accuracy when the pupil outline is well obtained. In many practical applications, however, pupil is extracted with defection and adjacent noise. The current study draws focus to devise a method of pupil center estimation for pupil outline image with defection and noise. The method consists of three steps. It firstly takes noise out using Hough transform and then defect part by analyzing outlier of curvature of the contour. The Hough transform is employed to find the center lastly. We compared the accuracy on finding the pupil center by least squares methods, Hough transform, and our method. Forty-five eye images were tested. The result indicates that mean error of least squares methods, Hough transform, and our method were 8.4, 6.5, and 2.4 pixels.