A novel method for eye features extraction

Extraction of the eye features, like the pupil center and radius, eye corners and eyelid contours from the frontal face images often provides us useful cues in many important applications, such as face recognition, facial expression recognition and 3D face modeling from 2D images. Assuming the rough eye window is known, the color information, Gabor features are extracted from the eye window and then these two features and their mutual relationship have been incorporated into the design of our algorithm. Firstly, after the pupil center is detected in H channel of HSV color space, the pupil radius is estimated and refined. Secondly, eye corners are localized using eye-corner filter based on Gabor feature space. Finally, eyelid curves are fitted by spline function. The new algorithm has been tested on our SJTU database and some other pictures taken from the web. The experimental results show that our algorithm has high accuracy and is robust under different conditions.