Face Detection Of Color Image Based On Segmentation Of Skin Color

Human face recognition is a research subject with a bright practical prospect and great challenges, which relates to some hot research fields such as pictures and images processing, neural network, pattern recognition and so on. It has grate significance of application in these fields such as public security tracing system, security-validating system, files-management system, credit card testing system, video conference system, and human-machine communication system.A complete human face recognition system is composed of human face detection, feature extraction, and match recognition, in which face detection is the first and the key step in face recognition. In the recent decade, face detection becomes more and more a hot point of research. In this thesis, a fast detection method based on the division of skin color in color images is proposed. First, some commonly color models in computer visual field are analyzed, considering the influence of eliminating lighteness, the YCrCb color space system is adopted, skin color model is established by using the characteristic of aggregation of the skin color, the skin colors and the background areas similar to the skin colors are divided, and a binary skin color map is generated. Then, using the images reshaping technology, the divided skin colors are filtered, the non-skin color noises in the connected face area are dislodged, and the connected area of the human face is obtained. Finally, the area selected whether includes human face should be measured by intensity and features distribution model, and the position of eyes and lips are lined out. A new method for fixing the position of eyes and lips is also proposed in this thesis. The human facecharacteristics such as the eyes, the lip are fixed position in the connected area, which is used to measure the existence of human face exactly, and lay the foundation for characteristics refinement, which is the next step of Face Recognition.According to the experiment results, the method proposed processes the robustness, and it could be use for face detection under complicated circumstances. The skin color model creation method based on YCrCb color space is researched, the result shows that it is a better skin color model; The human face identify method based on face characteristics is researched, the result shows that it is an excellent way to identify the real human face. Whereas, it has the limits, improving method is put forward against its shortcomings. This method could detect the faces of different sizes and expressions in the straight direction.