Thematic Indexing in Video Databases.
University of Southampton, ECS,
This thesis presents a novel video database system, called tv-DbMS, which caters for complex and long videos, such as documentaries and educational videos. As compared to relatively structured format videos like news or commercial advertisements, this database system has the capacity to work with long and relatively unstructured videos, by using the thematic indexing model. Thematic indexing is a novel way to track a story in a video. A video contains many themes, which are implicitly related to each other. In order to resolve queries about particular scenes in a video, the scenes are stored hierarchically, which provides “is-a” or “have-part” relations between them. In tv-DbMS video metadata is stored in a database. This metadata is used for indexing and querying the video. In this model, metadata contains information about segments (combinations of frames) and scenes (collections of segments that represent the same content). The metadata is organized in such a way that it can be used to navigate a theme (concept or idea). Annotations describing scenes are linked in a hierarchical manner to create a story line in the video, and thematic indexing is used to develop a video catalogue. Thematic indexing support the processing of complex queries in a video database and tracking story lines. Integrating open hypermedia capabilities, providing facilities for navigation, further enhances the tv-DbMS model. The model has been tested on documentaries made for television by the Earl Mountbatten of Burma. The results show that a user can easily query for natural scenes or events. The model is also tested on an educational video namely the inaugural lecture of Prof. W. Hall. The results show how queries on abstract topics discussed in the lecture, can be handled by using thematic indexing. The database system has been designed to ensure MPEG-7, and RDF compatibility. The metadata and other information are kept in such a format that it could be easily converted as captions to Audio Video Objects (AVOs).
Actions (login required)