Detection and localization of face feature area for man-machine interaction

In this paper, we describe methods of detecting a human face and localizing the feature area in our man-machine interaction project. We use a simplified human skin color model for pre-segmenting the input image. From candidate areas we use linear Fisher discriminate analysis (LFDA) to choose the most like face area, then in the detected face area, we localize the feature area with LFDA and linear principal component analysis (LPCA). We use the method of the image pyramid to improve the searching speed. The results are presented on images from the Chinese audio visual speech recognition database (CAVSRD 1.0).