Research On Facial Feature Points Location Based On Video Images

Facial feature points location is the key to face analysis techniques, it is widely used in face recognition, 3D face modeling and facial animation etc. However, it is very hard to locate facial feature points precisely due to the changes of light, gesture and facial expression and the occlusions of glasses and beard. In this dissertation, robust facial feature points location algorithm is studied to turning left and right human video sequence. The main points of research and innovation are as follows:1) The facial feature points are selected according to MPEG-4 facial definition parameters ( FDP) in the analysis of the major organs of human face. And the selected facial feature points are classified by the importance of follow-up applications and the ease of locating feature points.2) The stack ASM and two-dimensional contour lines of searching local feature points are introduced by improving the initial position of ASM and the searching strategies of local facial feature points. This method is used to locate the first frame facial feature points of video sequences which directly affects the positioning accuracy of the follow-up video facial feature points tracking.3) Different strategies to track the video facial feature points to the different facial gestures are proposed. Affine-corrected optical flow tracking method is used for small gestures of human face. Offset correction of the optical flow tracking method through the sub-region, calculating accurate points tracking for medium / large gestures of face. The human face images are converted to the HSV color space, using Sobel transform to extract the face contour is used to locate facial feature points on the outer contour. 4) The face gestures are estimated according to the inside corner of the eye points and the tip of nose as an application of facial feature point location by video sequences. Finally, the facial feature points location of the video sequence and estimation of face gestures have been implemented by Visual C++6.0 and OpenCV1.0.