Abstract: According to face detection for videos, a fast detection algorithm based on regional features was proposed in the paper.The comprehensive algorithm with temporal difference and background difference was used for acquiring and tracking the moving objects, which can avoid updating error of background model caused by moving object.In the region of moving objects, first Trisection image model of mosaic based on regional feature was used for detecting face region, then the frequency histogram method was adopted to combine detection area, and finally the face position was obtained.The experiment shows that the average detection time is 30 ms/frame and the accuracy rate obtains 95.7%.Therefore the algorithm which has lower complicacy and higher accuracy rate is suitable for real-time face detection for all kinds of video images.