Robust precise eye location under probabilistic framework

Eye feature location is an important step in automatic visual interpretation and human face recognition. In this paper, a novel approach for locating eye centers in face areas under probabilistic framework is devised. After grossly locating a face, we first find the areas which left and right eyes lies in. Then an appearance-based eye detector is used to detect the possible left and right eye separately. According to their probabilities, the candidates are subsampled to merge those in near positions. Finally, the remaining left and right eye candidates are paired; each possible eye pair is normalized and verified. According to their probabilities, the precise eye positions are decided. The experimental results demonstrate that our method can effectively cope with different eye variations and achieve better location performance on diverse test sets than some newly proposed methods. And the influence of precision of eye location on face recognition is also probed. The location of other face organs such as mouth and nose can be incorporated in the framework easily.