Automatic face region detection in MPEG video sequences

In this paper, we present a fast algorithm that automatically detects human face regions in MPEG video sequences. The algorithm takes the DCT coefficients of MPEG video frames as the input, and output the positions of the bounding rectangles of the detected face regions. The algorithm is divided into three stages, where chrominance, shape, and DCT frequency information are used respectively. By detecting faces directly in the DCT domain, there is no need to carry out the inverse DCT transform, so that the algorithm runs in real time. In our experiments, the algorithm detected 92% of the faces in one hundred MPEG I-frames with few false alarms. The algorithm can be applied to JPEG still images or motion JPEG videos as well.