A Rapid and Robust Method for Shot Boundary Detection and Classification in Uncompressed MPEG Video Sequences

Shot boundary and classification is the first and most important step for further analysis of video content. Shot transitions include abrupt changes and gradual changes. A rapid and robust method for shot boundary detection and classification in MPEG compressed sequences is proposed in this paper. We firstly only decode I frames partly in video sequences to generate DC images and then calculate the difference values of histogram of these DC images in order to detect roughly the shot boundary. Then, for abrupt change detection, shot boundary is precisely located by movement information of B frames. Shot gradual change is located by difference values of successive N I frames and classified by the alteration of the number of intra coding macroblocks (MBs) in P frames. All features such as the number of MBs in frames are extracted from uncompressed video sequences. Experiments have been done on the standard TRECVid video database and others to reveal the performance of the proposed method.