Conclusion and Extensions

Abstract

A system was proposed in this thesis for automatic segmentation, indexing and retrieval of audiovisual data based on multimodal media content analysis. The video stream was demultiplexed into different media types such as audio, image and caption. An index table was generated for each video clip by combining results from content analysis of these diverse media types. Structures for different video types were described, and models were built for each video type individually. This general modeling and structuring of video content parsing is very unique. It achieves more functions than existing approaches which normally adopt a single model with focus on the pictorial information alone.