Research on Technologies of Shot Boundary Detection and Key Frame Extraction for MPEG-4 Video

How to retrieve and browse valuable information quickly and accurately from mass data is a problem to be solved,that is hard to traditional text-based video retrieval system. Therefore, content-based video retrieval technology is raised, and gradually becomes a research hotspot in multimedia field.As the first step when processing video data, shot boundary detection and key frame extraction are the most basic and the most important parts in video retrieval system. The precision of boundary detection and Key frame extraction,makes a great difference in the results of video retrieval and browse. Therefore, they are critical in CBVR system.Nowadays, more and more video data are stored in compressed format such as MPEG-4. In this paper, research is focus on shot boundary detection and key frame technology for MPEG-4 compressed video. Traditional boundary detection and key frame extraction algorithm is based on non-compressed domain, which is inefficient because it handles extremely large amount of data and need much space. In view of this disadvantage, an efficient scheme for automatically detecting the shot boundary and extracting key frame in MPEG-4 video stream is proposed in this paper, which handles the compressed data directly without decompresses it. In the proposed scheme, dissimilarity between adjacent I-frames is firstly calculated by DC coefficient, to find out the frames sequence with shot change. Through comparison of the relative macroblocks’type,the precise place where shot change is then found out. Then,due to the fact that a shot may contain frame sequences of different active characteristic,an algorithm based on motion characteristic of frames is proposed. Finally, the results of the experiments demonstrate the correctness and effectiveness of the scheme proposed in this paper.