An Automatic Eye Detection Method for Gray Intensity Facial Images

Eyes are the most salient and stable features in the human face, and hence automatic extraction or detection of eyes is often considered as the most important step in many applications, such as face identification and recognition. This paper presents a method for eye detection of still grayscale images. The method is based on two facts: eye regions exhibit unpredictable local intensity, therefore entropy in eye regions is high and the center of eye (iris) is too dark circle (low intensity) compared to the neighboring regions. A score based on the entropy of eye and darkness of iris is used to detect eye center coordinates. Experimental results on two databases; namely, FERET with variations in views and BioID with variations in gaze directions and uncontrolled conditions show that the proposed method is robust against gaze direction, variations in views and variety of illumination. It can achieve a correct detection rate of 97.8% and 94.3% on a set containing 2500 images of FERET and BioID databases respectively. Moreover, in the cases with glasses and severe conditions, the performance is still acceptable.