Content-based Video Retrieval Research On Key Technologies

As computer networks and the multimedia technology develop rapidly, video data is encountered more frequently. Consquently efficient usage and resonable organization of video information turn out to be an urgent issue.Traditional text-based search methods can not meet the requirements of information processing. Thus content-based video retrieval technology (CBVR, Content-based Video Retrieval) was proposed, in which colors, textures, shapes and edges of objects are applied to describe video fragments.Shots detection, key-frame extraction and matching indexing are the core technologies of CBVR. In this thesis, based on existing algorithm we research and improve the above technologies as follows:Firstly, the current shots detection algorithms are analyzed. We suggest that the application of template matching algorithm on video streams composed by DC microfilms not only removes noises in images but also compensates the motion of small objects and shots. And it is shown in experiments that the detection rate and precision of this algorithm are satisfactory. An improved key frame extraction algorithm are proposed, which focuses on the center of images. The proposed algorithm extracts key frame more precisely and produces less redundancy compared to other algorithms.Secondly, an algorithm matches similarity based on colority local cumulative histogram combined with texture features is proposed, which searches more quickly and more precisely.Finally, a content-based video search system framework is designed according to overall requirements of search system, which is based on the algorithms proposed in the thesis and includes video preprocessing and video query subsystems.