Based On A Variety Of Information Of Dynamic Face Detection

The topic of face detection derives from that of face recognition, and has emerged as an individual important subject. This thesis is on the realization of a dynamic human face-detection system with very high hit-rate and slow false-alarm-rate. In this paper, a dynamic face-detection system based on haar-like features with the frame difference motion detecting front and the face-color detection post processing in combined color spaces is proposed and accomplished end.The proposed dynamic face detection procedure is divided into three phases:The first one is the adjacent frame based motion-detection, in which the morphology method is employed to extract the motion objects and their edges, returning the rectangular range of motion-part to the haar-like classifier. Because most of non-motion region is removed in this step, the operating speed of the next step is promoted.In the second phase, the concise haar-like feature is used and layers of classifiers are reduced according to the characteristics of the proposed system, which further speeds up the processing rate.In the third phase, the face-color detection in a combinatorial color space of the rgb and YIQ effectively decreases the false-alarm-rate of the system.Experimental results have shown the proposed system can attain 93% hit-rate for common used video sequences, which is higher than that of the normal haar-like feature based classifier. The false-alarm-rate of the proposed system is far lower than that of the haar-like feature based one. The speed of face-detection in the system is 11 frames per second, meeting the requirement.