| As the most complex media form of multimedia, video is applied and developed widely due to its diversiform representation, abundant semantic content, and convenient ways of recording. On the other hand, the development in computing technologies, broadband communication networks, and mass storage devices have resulted in large amount of video data being generated and made accessible in digital form throughout the world. As a result, the main tasks of video information processing are effective video content analysis and efficient video retrieval.In this dissertation, a video retrieval method based on low level features is proposed. It settles the problems of fast video clip retrieval in large amount video database with effective organization of video features database by VA-File and fast query algorithms by restricted sliding window. In addition, we design a new similarity measure which synthetically considers the visual similarity and temporal order between video clips.Since ontology can be used to express the semantic relationship between concepts in videos, we propose a video semantic information extraction algorithm based on ontology for video content analysis and retrieval. Two different methods of Bayesian method and Bayesian Network are used to construct video semantic ontology, and reasoning rules based on ontology are applied to detect the abstract concepts in terms of concrete concepts in videos.In allusion to the complex requirement of query, a new video retrieval model based on multimodal information fusion is brought forward in this thesis. It includes multi-models like text, image, high-level semantic features extraction etc. and uses relational algebra expression to fuse multimodal information attained by multi-models retrieval. Experimental results demonstrate that our method could manifest the advantages of multimodal information fusion based on relational expression in video retrieval, and achieves good performance on complex semantic video information retrieval.
|
PDF Full Text Request