Automatic Shot Boundary Detection Combining Color, Edge, and Motion Features of Adjacent Frames

This paper describes the contribution of the Center for Computing Technologies (TZI), University of Bremen, Germany, to the shot detection task of the TREC 2004 video analysis track (TRECVID). The approach uses RGB histogram values which are calculated within a five frames width window and the edge change ratio between consecutive frames as well as frames at a distance of 10. Both methods are used to detect hard cut candidates. To confirm or reject the hard cut candidates a block-based motion analysis is used. Gradual transitions are also detected using RGB histogram values together with a finite state machine.