Video Shot Boundary Detection Based On MPEG Compressed Domain

Video shot boundary detection as the first step of the content based video retrieval is a very important step. The quality of the video shot retrieval will directly relation to a series of the subsequent works:key frame extraction, feature extraction, scene segmentation, so shot boundary detection is a challenging research direction.In summary, this article conducted in-depth study of video shot boundary detection.First, this article discusses the background and significance of the research topic, the research status at home and abroad, the existing problems, then summarizes the comparison of the advantages and disadvantages of uncompressed domain and compressed domain video shot boundary detection classical algorithm.Secondly, this paper will be mainly research the MPEG compressed domain video shot boundary detection. Therefore, the main contents of the third chapter is the MPEG video code technology, the structure of the stream level analytical. Finally, this paper based on the MPEG-2standard C Model with VC6.0completed a stream analyzer, this stream analyzer can extract the compressed video sequence section feature information.Thirdly, this article describes the MPEG video compression standards, learning and summarize the existing uncompressed domain and compressed domain shot boundary detection algorithm, then proposed a shot segmentation algorithm. The first algorithm of this article is:matches the SIFT key points of the adjacent Ⅰ-frame DC figure, statistics the similarity and poor similarity of the P frame and B frame macro-block type information, to achieve the shot boundary detection.Finally, the second algorithm uses the compressed stream in I frame DC figure and P frame motion vectors information to detect shot boundary. This method is no need to distinguish the video campaign is mainly caused by the camera as a whole movement or by the local movement of the object and only need to analysis the dominant component of the relative motion.The experimental platforms of the proposed algorithms is VC6.0and Matlab. TRECVID videos and the videos downloaded from the internet are used to verify the proposed algorithms all have a good detection result.