The Research On The Facial Motion Analysis In Sequence Images

The face detection and tracking based on video is one of the key problems in computervision and pattern recognition. In the video surveillance, human-computer interaction, videoconferencing, identity authentication and multimedia field, it has a wide range of applications.Human face processing in video is divided into three steps: face detection, face tracking, andface recognition. Face detection and tracking are often more difficult because of complexbackground environment, change of outside light and complex facial gestures. In this paper, howto effectively solve the above problems and improve robustness and accuracy of tracking is thefocus research.This paper mainly studies the face detection and real-time tracking in the complexenvironment. Two problems of real-time face outline extraction in dynamic images are asfollows:(1) How to detect the real-time face region in the complex and continuous image.(2)How to the track movement face outline in the subsequent images.Aim at these problems，this paper combines the skin model and the difference movementmodel to extract the initial position of the face outline, then the face exact position is extractedusing two active contour model algorithms, the Level Set and the GVF Snake, and the strengthsand weaknesses of the two methods are compared. In order to develop the tracking algorithmreal-time in the dynamic image sequence and to solve the problem of the blocking, the facetracking algorithm based on GM(1,1) is put forward. According to the continuity of themovement of the face outline, the iteration of the active contour model is based on the faceoutline movement rule which is given out by the mass displacement of the face outlineforecasted by using the GM(1,1) model. At the same time the center mass of the face outlinefrom the active contour model algorithm is taken as the basis for the next image GM(1,1)projection.The GVF Snake algorithm can’t extract the edge accurately when the track human faces isfar to near. Besides the GVF Snake model is difficult to solve the problem of deep depression. Inorder to solve these problems and make the occlusion problem have a better result，this paper proposes a face contour extraction method based on the improved GVF Snake combined withMean Shift. The Mean Shift is first used to get the initial face region and then the improvedGVF Snake algorithm is to extract the accurate face contour to ensure an effective solution ofGVF Snake algorithm initialization and the deep depression which leads the variable can not tobe iterated. The Mean Shift and GM(1,1) can solve the blocked problem effectively. Andexperimental results show that this algorithm is effective.