Face Detection And Tracking In Video Sequences

Along with the rapid development of information technology and the needs of personal identification, security inspection, expression analysis and intelligent human-computer interaction system, the research on human face problems has much more practical values and research significance as an important branch of computer domain. Human face problems consist of four parts: human face detection, human face tracking, human face recognition and the derived analysis of pose and expression. This paper does research mainly on human face detection and human face tracking.In terms of human face detection, based on the achievements of previous researchers, some work have been done on the problems of complex lighting condition, pose-varied and dealing with sheltering in this paper as below:1. In order to extract skin-color area much precisely, it corrects the color bias based on a lighting compensation technique that adjusts the color-histogram of picture linearly by the methods of"reference white".2. Extract skin-color area by skin-color model and turn the area to binary image, and then get the candidate as human face image after closing operation and merging the nearby area. In this way it gets rid of most background and reduces the detecting area.3. In color image, a method for human face detection is done based on the character of color. First of all, it extracts the position of eyes and lip because of the difference between eyes and lip and skins in color. And then it makes use of the position relationship of approximately isosceles triangle between eyes and mouth to make the rules for final face decision.4. According to the problem that lighting condition affects the result of extracting skin-color area seriously; an algorithm for face detection of gray picture is presented in this paper which integrates a series of weak classifier based on the face character of gray distributing, arithmetic of Principal Component Analysis and morphological processing. It traverses the whole picture with a small window. Firstly, get rid of some non-face windows with the face character of gray distributing. Secondly, test the windows with a weak classifier using the arithmetic of Principal Component Analysis. Thirdly, deal with the parts of eyes in a series of ways such as calculating vertical grads, threshed, dilate, erosion and so on. Then we can make the decision of whether the window is whether a face or not based on the information of eye's shape and position. If it is, get the position of eyes. The final test results make out that this algorithm takes good effects in different lighting condition, face size, expression, pose and in partial occlusion and in complex background.In terms of human face tracking, a face tracking algorithm under complex background based on particle filter is presented in this paper for the problems of occlusion, variances of lighting and background, the non-rigid character of human face and the non-regular and non- proportioned.The algorithm initializes a face template of proportioned color-histogram firstly, and then matches the template in particle space and calculates the similarity, finally gets the tracking position of face. Particle filter can deal with those state estimation problems of nonlinear and non-Gaussian without the measurement equation which is hard to build in face tracking. Experiments results demonstrate that the algorithm is effective under complex situation such as changing background and partial occlusion.At end of this paper, a human face detection and tracking system is constructed in the environment of VC++6.0 for the use of detecting and tracking face in video. The flow chart of the system is also given in the end.