Real Time Face Feature Detection System Based On Video Image

Human face detection roots from face recognition, It is the key sector of auto human face recognition. In the early stage, human face research is largely based on restrict human face image(ie. Images without background),and human faces are supposed to be obtained easily. So human face detection had not been emphasized greatly. Human face feature detection is based on face detection, the feature detection is detected on the detected face area.In this paper, the illumination and the background are easily accommodated, and most people are easily to be suited in the human feature detection system. We adapt the haar feature detection algorithm to detect human faces and eye rectangles and use a new method to detect pupil and inner eye corners. The detection precision is much higher and detection adaptability is largely improved. First we adapt the haar feature detection algorithm to detect the human faces, we detect the eye, mouth and nose rectangle on this foundation. Then, we adapt the improved iterative algorithm to detect the pupil center and adapt the corner detection algorithm to detect the inner eye corners. At last, we adapt the skin color algorithm to get the precise face boundaries. We also detect the eye blink as the mouse click event.In order to improve the human feature detection precision and the eye control mouse precision, the system adapt the big pixel image as the processed image. The big pixel image is used to locate the pupil center and the inner eye corners. Big pixel image is translated to the small pixel image to detect the face and eye rectangle, then we get the eye location on the big pixel, so that the big pixel pupil and inner corners are detected by some algorithm. So that the system's real-time is improved.Also, a frame prediction mechanism is proposed for improving the real-time system in the processing of human facial features detection, namely the localizations of facial features of current frame are estimated approximately based on the latest already- detected frame, thus reducing computation time of detecting and tracking the target, improving the system's real-time.