The Key Technology Research Of Content-Based News Video Retrieval

With the rapid development of data communication technology, compression technology, software and hardware technology, transmission and storage of vast video information has made it possible to achieve. Video information gets rid of normal restrictions, and plays more and more role in people life and work. In order to achieve efficient organization, analysis, management and retrieval, the Content-Based Video Retrieval (CBVR) system emerges as the times require. The CBVR is becoming an important object in the area of information science. The news video deals with wide contents and gets much more attention than any others, moreover, it gets special structure and organizational characteristics. For this reasons, the research based on news video is becoming a central issue in the area of CBVR.Our research is based on the background of CBVR. We introduce some basic concepts, system structure and key technology. We analysis the methods of shot boundary detection, key frame extraction and shot clustering in video processing. And then we get some improved algorithms that are based on the information entropy, histogram information and clustering algorithm. The theoretical analysis and experimental results witness that our algorithms are efficacious. In the end we design a news video processing system about the key technology.The main tasks of this dissertation:(1)In order to achieve efficient video organization and retrieval, we get research on the shot boundary detection. We propose a shot boundary detection which is based on computing the mutual information for histogram that is calculated by HSV color model. The experimental results witness that our algorithm is an efficient shot clustering algorithm. (2)We get research on the key frame extraction, information entropy and histogram information. We propose an approach of key frame extraction based on mutual information and clustering algorithm. The experimental results witness that the algorithm is an efficient algorithm for news video shots.(3) This dissertation proposed a simulated annealing K-means clustering algorithm for video shots. Then we used the global optimize ability of simulated annealing algorithm to remedy the local extremum shortcoming of K-means, so that we can increase the clustering accuracy. The theoretical analysis and experimental results witness that our algorithm is an efficient shot clustering algorithm.