Content-based News Video Retrieval

With the development of the Internet and digital media technology and the emergence of mass digital video, we need to facilitate access to a large number of video information in video information of interest, Thus, content-based video retrieval(CBVR) become an important issue in the field of computer science. This paper choose news videos as its research object and is based news videos'features of special structure,organization and retrieval way,which make it one of the most important topics。We have do research as follow.(1) This put forward a improved the shot boundary detection algorithm. Upon completion of the structure analysis of video and key technology research work, compared method of compares the pixel, edge profile characteristic method, gray level difference method, and dual histogram histogram shot detection method, We propose an improved method for shot boundary detection, based the complementary characteristics of pixel comparison, the histogram method and double histogram.(2) In news feature extraction part of this paper, a fixed area template matching shot detection for news video host is proposed. First of all standards host lens is detected by EFR Methods and thus determine the video search area, with block HSV color histogram as a template parameter, on behalf of all the possible lens frame template matching to achieve the ultimate host Lens confirmed. This approach can guarantee the detection of the flexibility of the system can reduce the false detection rate.(3) Based on the shot segmentation, shot detection of host, theme subtitle detection, advertising video scene segmentation, audio streams semantic extraction of the division, we determine the completion of the video scene. For video indexing, browsing to provide the organizational framework, the semantic information automatically generated based feature extraction, can quickly point users to find video clips with better efficiency. This comprehensive utilization of features of video and audio and its own model processing of news specific, makes the process steps to clear, the processing results to objective and accurate. With the current outstanding research achievements, this paper can realize content-based video retrieval better.