A robust method for eye features extraction on color image

This paper presents a robust algorithm to extract eye features, including pupil center and radius, eye corners and eyelid contours, from frontal face images. These features are very useful cues for applications like face recognition, facial expression recognition and 3D face modeling from 2D images. With the assumption that the rough eye window is known, the proposed method integrates color information, Gabor features and the mutual localization relationship between different features. Different from other methods focusing on gray scale, our method detects and estimates pupil center in H channel of HSV color space, and then pupil radius is estimated and refined. For eye-corner localization, a Gabor eye-corner filter is constructed to detect the corner point. It is more robust than projection methods or edge detection methods. Based on the previous results, eyelid curves are simply fitted by spline function. The proposed algorithm has been tested on our SJTU database which contains more than 700 pictures. The experiment results show very good robustness and accuracy.