Research On Fast And Applicable Face Detection And Tracking Algorithms

Face detection is to detect human faces and provide the exact coordinate of each face in still images or video sequences, regardless of different location, orientation, size, pose, lighting condition. As a key technology in human face processing, face detection is of great importance in the field of security protection, ROI-based coding, content-based image retrieval, automatic video surveillance, human computer interface, etc. Now, face detection is one of the most active research fields of pattern recognition and computer vision. With the great efforts taken by researchers all over the world, face detection can now achieve a usable detection rate and speed. However, human face is a nature structure with highly complicated variations in detail, which bring great challenges to the performance of detection algorithm. These kinds of variations lie in pose, facial expression, partial occlusion, lighting condition, rotation, etc. Besides, human faces are always compounded with a complex background. Due to all of these difficulties, there is no such an algorithm that can handle all these variations without any kind of limitation at present. In this thesis, main research works are focused on the general problems in face detection/tracking and try to provide solutions to these. The objective is to build a complete face detection/tracking algorithm in order to satisfy the requirement of practical applications. The main task lies in the following four aspects: (1) Efficient skin segmentation pre-processing algorithm. An efficient skin model based on CrCbCg color space and fuzzy cluster is constructed, and can be used effectively in pre-processing phase. (2) Fast, accurate detection algorithm. A core classifier based on Haar-like feature and AdaBoost learning algorithm is built, and can provide high performance in the core phase of face detection. (3) Multi-pose human face handler. The handler is based on radial template and parallel connected classifier to deal with the in-plane and out-plane rotation of human face, which can broaden the useful field of detection algorithm. (4) Combined with surveillance system, make it more applicable. Apply the face tracking algorithm to actual video surveillance system and bring the intelligence to common surveillance system. Plenty of experiment results and comparison with other published method show...