Face Detection In Video

In recent years, with the quick development of computing technology, the use of digital video becomes more and more popular. Faced with the emergence of video data, how to find the necessary information on the video is an urgent problem. The research of face image retrieval on video information is a hot and hard point of computer graphics and computer vision. Face detection and tracking is an important research aspect in artificial intelligence and computer vision. As a key technology of face information processing, it has a broad application values in many fields such as content-based image retrieval, visual surveillance, etc.In this thesis, a study on face detection in video is presented. Because of the complicated pattern and the frangibility of human face, most face detection algorithms have the weakness of large computation, low efficiency and low robustness. Color detection algorithm is simple and rapid, but cannot overcome the influence of illumination and background color disturbance. In 2001, Viola present a fast face detection method based on AdaBoost learning algorithm, which is one of the most advanced face detection algorithm. Although it is very real-time, but under the real scene, face detection rate and so on need to be further improved. After analysis of the advantages and disadvantages of existing face detection algorithm, this paper proposes to adopt AdaBoost algorithm to localize face region at first. The experiment indicate that the AdaBoost algorithm greatly enhances the examination rate of face detection. Then, through the establishment of skin color model based on the YCbCr space, we map the early detection of the face region to YCbCr space and count the percentage of skin color spot in the area to determine wheather it is face region or not. As the calculation of the optical flow of the entire face is very time-consuming, We can get the head Harris corner points. Finally, we can use the optical flow change of the head Harris corner points to make the estimate of the face movement tendency in the sequence picture.This thesis uses VC++6.0 and OPENCV to develop face detection program based on the video. Through the experiment, we basically realize the face detection on video and reach higher detection rate and lower leak detection. The experimental results show the system has good real-time performance and accuracy.