Face Detection Based On Skin Color Model And Template Matching

Human face which contains a wealth of information, is a typical feature of biological, and has broad application prospects in areas of personal identification, video surveillance and human computer interaction intelligent. With the development of artificial intelligence, machine vision and pattern recognition, the applications of face detection which based on facial features become more and more important, has aroused the concerned by majority of researchers, and gradually developing into a relatively independent research. The speed and accuracy of face detection is the goal for the algorithm.Face detection which is to determine all the numbers, location, size and other information of the face (if present) which in the input color image. Since the color images are obtained through most of imaging device, the face detection which based on color images has become a research hotspot. The skin color is an important feature to face, it does not rely on human particulars of facial and gestures, and has relative stability, and the template matching can precisely located the matching position. So this paper studies the algorithm of face detection in two aspects, the specific work as follows:(1) The difficulties of the face detection caused by the face tilt, illumination and others which existence in face images, using the YCbCr color space which has good characteristics in color clustering, and propose the algorithm of face detection which based on Gaussian model, and calculate likelihood that the image pixel belong to the skin color through the skin model, get the skin color likelihood image. At the same time, reduce the impact by the interference of light and elimination impact by the head rotation for face detection.(2) Proposed a method which based on Fisher criterion could segmented the skin color and non-skin color image through selected the threshold adaptive, and then we detect and segment the skin color likelihood image by this threshold, improve the effectiveness of skin detection. On this basis, this paper which according to the geometric features of face, screen the face region by some rules, obtaining the rough location of face region. (3) In order to locate the face position accurately, this paper proposed the algorithms of face localization which based on template matching. First of all we established the average face template, removed the interference of irrelevant information through the mask, and then calculated the correlation in the face image by the template changed self-adapting, improved the accuracy of face detection.Finally, the algorithms in this paper are implemented by the MATLAB7.7 programming, and be tested through a large number of images. The results of simulation experiments show that: it can effectively extract the skin color regions through the skin color Gaussian mode in this paper. In the skin region, the paper screening the candidate region effectively according to the geometric features of face, play the role of coarse positioning. Finally we get the precise location of face through the average face templates. Simulation results show that the algorithm is effective.