Eye Location Based on Hough Transform and Direct Least Square Ellipse Fitting

Feature extraction and location of eye are very important in face expression recognition, human-computer interaction, face detection and identification etc. This paper presents a new efficient method for eye location. Firstly, gradient, integral image, and Hough transform are used to obtain location and radius of the pupil. Secondly, CANNY operator and ellipse fitting of direct least square are used to obtain elliptic equation of the eyelid. Finally, the center of the ellipse is deemed to be the center of the eyelid, and the edge of fitting ellipse is deemed to be the edge of the eyelid. This method is extremely robust and accurate in the condition that two eyes are horizontal and inclined. The average error is less than 1.82 pixels.