Study Of Content-based News Video Retrieval System

With the development of the internet and the video compression technique, and the increase of network digital videos, it is possible to retrieve and store a great deal of video information. Retrieval is the essential technique for organizing and analyzing information efficiently. Thus, content-based video retrieval (CBVR) is an important issue in the field of computer science. Among various videos, news video has its own special structure, and ways of management and retrieval, which make it one of the most important topics.Considering the unique structure of news video, this paper presents a semantic scene segmentation model taking advantage of both video and audio information. Using SQL Server2000, a massive capability database is implemented for saving video data with a general news video retrieval system.In the scene segmentation model, shot detection, key-frame extraction, scene-change detection, anchorperson detection and speech recognition are fused to realized the semantic analysis and structured processing of the news video, which is saved in the database to form a video database. Then, an Web-based B/S retrieval system is constructed with text and image.The experimental results illustrate that the modularization of the algorithm greatly reduces the complexity of retrieval system. It makes the processing steps more clear and concise, the results more objective and accuracy with the combination of both video and audio features. Various ways of retrieval meet the customers' need better. It is proved that the system has good practicability.