Study On Method Of Video Retrieval Based On Shot

With the arrival of information and network times, a great deal of visual information is collected, transferred and applied all over the world. However, the problem that the information increases rapidly has been very serious. In many fields a lot of visual resource collected is set aside since it is not processed, which brings about large waste of resource. Besides, the limitation of traditional video information management system based on text has been revealed. How to use, express, manage, inquiry and retrieval video information has become urgent, and then the technique of video retrieval based on content arises. In this thesis, video retrieval based on shots is researched, which takes video shots as fundamental analysis elements, and finds the satisfactory shots according to similarity rules in the shot database.The video retrieval based on shots is studied. The content of shot database is organized according to shot clusering and the system structure of video retrieval based on shots is proposed. Not only the complexity of similarity calculation could be decreased but also the losing of similar shots could be avoided effectively, and the retrieval speed is also improved.When the retrieval process is realized, the content of video frames is expressed by combining color with space characteristics. Then the algorithm of key frame extraction based on clustering of optimizing initial centers is adopted. The initial clustering centers are determined based on the distributing of video data, so the stability of results could be strengthened and the extracted key frames are able to represent the visual content more effectively. The condensed clustering method is adopted to cluster shots, which can realize the shot clustering automatically and accurately and don't need to divide the number of cluster in advance. By this clustering method calculation of shot similarity can be avoided one by one effectively.Two calculation of similarity is used. Firstly, the large threshold is set to assure the retrieval range. Then Hausdorff distance is used to measure further, and the result is output until the user is satisfied. This method takes account of both the similarity of shots themselves and the concrete demands of users, which strengthen the practicability and effectiveness.Finally, the retrieval experiments are carried out based on different kinds of inquiry shots and the results are compared. It ls proved that the effectiveness and accuracy of the method is improved greatly.