A method of eye center localization with Harr-like AdaBoost cascade classifier and Isophtoe features

Human eye center location plays an important role in the research and application of computer vision, involving tracking, human-machine interaction, emotion recognition etc. For detecting the eye center location conveniently, a method is proposed based on Haar-like feature and isophote features without complex training process. Firstly, the Haar-like features are employed to detect faces by AdaBoost cascade classifier. Then, the exact eye region is segmented based on the detected face region by using image corrosion method. Finally, the isophote curvature feature is extracted to obtain the eye center. Experiments are conducted in three databases, including JAFFE, FERET, and ORL. The experimental results demonstrate that this method is feasible. To some extent, it has stability to detect the eye centers in the vertical direction.