This paper presents a method to determine the ordering of feature (attention) points for face detection. The method is based on the cross validation. The order is determined by the error rates of the face and non-face classification for test samples. It is confirmed that high discrimination rate is obtained by checking in the order of the selected feature points. In this paper, the weighted matching by information and Bayes estimation based method are used as face detector. By ordering of the feature points, we can reject the input image as non-face by checking only a small number of feature points. This can speed up the face detection.
展开▼