The Research Of Content-Based Video Media Information Retrieval Method And Frame

With the rapid development of the technology of computer, network , compression of video and audio and computer hardware support, it is no longer a severe problem to store and transmit the data of video media. Furthermore, digital video is widely used in people's life. However, the problem that video media information increases rapidly has been very serious. In many areas, a lot of video information collected is lay aside since it is not processed, which causes large waste of resource. In addition, the limitations of traditional video retrieval system based on text have been revealed. How to effectively organize, express, store, manage, quick search and browse video media data has became urgent needs of video field. Therefore, the content-based video retrieval system arises.This paper aims at deep analysis of content-based video retrieval from three aspects: the shot detection, semantic classification, and multi-modal information fusion.As for the shot detection, we will discuss the shot detection algorithm ,which is based on the histogram entropy difference. Use continuous inter-frame entropy difference to detect cut transition and interval inter-frame entropy difference to expand the gradual transition. Combine gauss model with slide window method to decide the adaptive threshold, and use the subsection best threshold decided by maximum entropy segmenting to eliminate the fault detection owing to the local maximum of slide window. Histogram entropy difference, as a shot detection, can efficiently expand the contrast effect between shot and non-shot frame. It is found in experiments that this method has a strong inhibition on interference such as flashlights, mosaics, etc.For clustering index and semantic classification, this paper discussed Bayes-based semantic video classification method through the clustering index of semantic-based supervision.In allusion to the complex requirement of query, a new video retrieval frame based on multimodal information fusion is brought forward in this thesis. It includes multi-modal features such as text, image, high-level semantic concepts, etc. The framework takes full account of the multi-modal characteristics of the text, images.