Research On Pragmatizing The Head-Shoulders Video Sequence Coding Algorithm

With the development of multimedia communication, veiy low bitrate coding has become the forefront of digital image compression. Under the guidance of the content-based compression, the author assisted Kongqiao Wang, a Ph.D. candidate then, in proposing and developing a very low bitrate, object-oriented image coding method, which is called the Head-Shoulders Video Sequence Coding Algorithm. This method is used to compress the large quantity of head-shoulders video sequences in applications such as videophone, videoconference and human-computer interface, etc. It consists of the human face detection, human face tracking, and scalable wavelet image coding algorithm. However, the performance of this algorithm is still far from the practical application, especially in the facets such as real time and robustness, etc. Therefore, the author has pragmatized the Head-Shoulders Video Sequence Coding Algorithm independently. The main achievements are as follows. 1. The speed of the human face detection algorithm has been greatly improved by the measures such as redesigning the method of decreasing the resolution, i.e. replacing the mosaic method with smooth-subsample method, adjusting the sfructure of the algorithm to eliminate its redundancy, and optimizing the programming, etc. Thus, the problem of real time of the human face detection has been solved to some degree. 2. The human face tracking algorithm has been greatly speeded up and its problem of real time has been successfully solved by adjusting its optimization order, i.e. replacing the search order of translation梤otation angle with rotation angle梩ranslation. Meanwhile, the measures such as noise elimination, analyzing the acute motion of the human face and so on, have basically solved the problem of robustness. 3. Under the premise of very low bitrate, the problem of real time of the coding has been elementarily solved, by adopting the method of block DCT transform coding along with unitary motion compensation. 4. A real time testing system has been established on LAN and PSTN to integrate and test the pragmatized algorithms. Besides the pragmatizing research, the human face detection algorithm has been extended from detecting faces in the gray-level image to detecting faces in the color image by introducing the color segmentation. And based on this work, a new MPEG-7 proposal has been submitted to the MPEG committee.