The Combination Of Pixel Wise Clustering And AdaBoost Algorithm For Face Tracking

Face detection and tracking is a fundamental problem in computer vision. Detection of faces is a critical part of face recognition, face expression recognition and critical for systems which interact with users visually. The face is the most distinctive and widely used key to a person's identity.Recently, The problem of detecting facial parts in video sequences is beyond the area of face information research, and becomes a popular field of research due to emerging applications in identity clarification, content-based image search and retrieval, accessing security, surveillance systems, intelligent human-computer interface, video conferencing, image database management system and so on.This thesis mainly introduces some solutions to the face detection domain. In our system, two kinds of algorithm are used to mutual assistance. One algorithm is based on AdaBoost for detecting frontal faces in still image and tracking frontal faces in video sequences. A skin tone detecter trained by AdaBoost and a filter built by the information between frames is first used to estimate the region of any potential faces in the video sequences. The potential region is then detected by the Attentional Cascade which is trained by AdaBoost. Well then, tracking frontal faces in sequences is realized. The other algorithm is based on K-means clustering for human head back tracking. If the face turning back to the camera, tracking would be dropped down since Cascade Classifier can't locate head back. Here, object tracking algorithm is proposed to track the head back when this exception occurs. Tentatively, we use the system combining with the two algorithms to solve the problem of frontal face and head back tracking in video sequences satisfactorily.The system has been evaluated on color video sequences, which contain frontal faces and head back against cluttered backgrounds. The face in the video sequences should be turning quickly, without long time obstacle. We use some video sequences to illustrate the robust property of the face tracking system.