Human Face Detection Based On Color Model And Bayesian Discriminating Feature

The technology of biologic character recognition is wildly applied in finance field, security field, and so on. The face recognition is more credible than other recognition mode, and the face detection is the first step in face recognition. In this paper, the front face detection of color still images is modeled by mathematic method. Firstly, Skin information is used to reduce the search range firstly. Then the exact face location is detected by using the ellipse template and the face intensity information, respectively. In skin detection, the K-L transformation method within the RGB color space, the YCbCr color space respectively and the clustering method within the YCbCr color space and the HSI color space respectively are compared. Then according to the compared results the skin clustering method is improved using the subsection clustering. When detecting the face with the ellipse template, the edge detecting processing is performed firstly, and the consistence of the gradient direction of the edge point and the tangent direction of the corresponding point in the ellipse template is used to decide if there exists a face. The Eigenfaces method and wavelet based bayes classifier method are also compared using the intensity information on face database and nonface database, and the experiments shows the wavelet based bayes classifier method can obtain a better results.