Research Of Face Detection Based On Multi-feature Fusion

Skin color is a very important biological feature of the human body. Skin color segmentation is widely used in face detection, web image filtering, face image retrieval, sex classification, age discrimination, medical diagnosis, video surveillance, face recognition, identity verification and so on.To the problems of current face detection, the adhesion problem between face and other regions where the color value is similar to the skin, and the face detection rate of multi-pose is relatively low. This paper does some in-depth studies, the main content includes:Firstly, the skin color segmentation algorithm based on Multi-gaussian and Bayes. This method through Multi-gaussian to fit the probabilities of chroma of cropped skin color images and natural pictures. We analyze the distribution of the two prior probabilities and use the decision theory of Bayes to segment color images. Compared with other commonly used methods, especially under the circumstances of the skin color like background very much, the segmentation of this method is more accurately.Secondly, the method of segmenting the problem of region adhesion based on skeleton and distance transformation. This method through extracting the vertex points of skeleton and some contour points as the feature points, calculating the distance transformation between the feature points and skin color regions, obtaining maximum points, according to the distance between the maximum points, merging maximum points and segmenting the area. Experimental results show that this method can remove some small background regions effectively, segment the region adhesion between the faces and other regions.Thirdly, face detection based on Multi-feature fusion. This method merges skin color segmentation, the same direction of features areas of eyes, mouth and eye brows and etc, and the location constraints of the center points of feature areas, employing multi-feature fusion to realize face detection. Experimental results show that this method can detect multi-pose faces more accurately.Lastly, the method of face contour smoothing based on distance histogram. This method uses the contour points and their center to form a distance histogram, and make use of the peaks and troughs of histogram to smooth contour. Experimental results show that this method can smooth face contour and reduce small background feature areas that are nearby the contour, so we can position the facial feature more accurately.