Research Of Human Face Detection Under The Color Background

With the development of intellectualized information technology, fast and effective Authentication is need in video monitoring, remote education, interactive technology, safety areas. Face detection technology is a biometric identification technology which identifies the status automatically according to the features of human faces. Face detection has lots of advantages among various biometric identification technologies, especially the advantages of visual and non-invasive, which make it has a broad prospect of application. It also draws the attention of academia and the industrial sector and has become a top subject. Many researchers are attracted by it and they have gained some meaningful achievements.First, the classification and research Status of face detection are introduced. Various color spaces are analysis and compared. Then YCbCr space is selected as the very space which is used in this thesis to build the skin color model. Assuming that the skin color is just related with color weight(CbCr) rather than light weight(Y) in YCbCr space and drop light weight when color space is mapped has some defects. The results of experiments show that only between the high light and low light area is color weight not related with light weight rather than in high light or low light area. As a matter of fact, the affect of light weight(Y) has to be taken into consideration if more accurate results are needed. Nonlinear color transformation is made according to the reasons above in order to make the skin color clustering in YCb'Cr' space. After that, skin model is used for segmenting regions in which may have faces. Then, line code is used for drawing the eigenvector of area from the images which are segmented. At last, FCM dynamic clustering method and ISODATA algorithm are used for detecting the areas of human faces from complex background. The program of reading, displaying, pre-progress, color space transformation, image segmentation, morphological processing, area label and extracting eigenvector from the images is realized on VC platform. The experimental result of segmentation by means of the new algorithm is satisfactory. This thesis includes several achievements as following:1)Some images may be affected by colorful lights, different sensitivity, enhancement factor or offset and the trichromatic uneven of images will come out. The color of the object will deviate its real one because of it. Color-balance is used to deal with it before detecting the skin color pixels. High light and shadow areas which are resulted from the environment or weather are also dealt.2) Line code are used when getting the features. of binary images, including the information of area, heart-shaped coordinate of connect regions. Then Morphology prior knowledge and Zernike moment invariants are used to normalize the feature vector so that the candidate areas can be confirmed.3) After the feature vector being extracted and normalized by line code and moment invariants, FCM dynamic clustering method and ISODATA method are used to identify the target.4.The current normal face detection technologies are analyzed. The face detection system is realized both in algorithm and VC platform.