Shot Boundary Detection Using the Moving Query Window

The large volume of video content generated each day has led to the need for efficient and effective methods of video indexing and retrieval. A common first step in indexing video content is to identify visually and semantically continuous segments or shots. In this paper, we present the moving query window approach to video shot boundary detection. This uses the techniques of queryby-example (QBE) and ranked results, both often used in content-based image retrieval (CBIR). Each frame of the video is used in turn as an example query on the image collection formed by the other frames within a moving window. Transitions are detected by monitoring the relative ranks of these frames in the results list. We show that this is an effective approach for the shot boundary detection task of the TREC-11 video track.