Structured representation and automatic indexing of movie information content

The development of video applications for digital multimedia has highlighted the need for indexing tools, enabling the access to meaningful segments of video. The high cost of manual indexing creates a demand for the development of automatic algorithms, able to extract such indices with little intervention. In this paper we present new editing model-based algorithms that automatically extract low-level features in a movie: camera shots and camera motion. Rules of him making are used to derive higher-level elements, such as shot-reverse shot sequences. The algorithms have been tested on 20 h of movies and comparison with techniques in the literature is provided. (C) 1998 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved.