An efficient gradual scene boundary detection algorithm for H.264/AVC in compressed domain

Scene boundary detection is the essential step for video analysis such as retrieval, indexing, browsing and so on. There have been quite a lot of scene boundary detection methods for MPEG video streams and they work well. However, the variable size of macroblocks and more type information of macroblocks in H.264/AVC bitstreams make it different from MPEG bitstreams. Consequently, most of the algorithms can not be applied to the H.264/AVC directly. In this paper, an algorithm based on the percentage of blocks coded with the same mode in each frame is proposed. After analyzing the statistics of type information of the sequence contain gradual scene changes, the algorithm sets appropriate thresholds to locate the gradual scene boundaries. Experimental results show that the proposed algorithm performs well and can detect 90.9% of the boundaries and make few false alarms.