Research In Automated Video Segmentation And Content-Based Classification

Content-based video retrieval system has been one of the research focuses in recent years. The so-called "content-based video retrieval" corresponded with the past retrieval based on whole file is on the basis of the video data and correlates with the content. This needs to carry on video analysis and segmentation.The news program occupies the important proportion in resources of the video. It is made up by the news section called news story unit which has relatively independent content. This subject uses the news video of local TV stations as an example, aims to implement automated news story unit detection, has carried on the following research mainly: first, in the condition of characteristic of the video captions and a large number of appearances which interfere the captions, " Part Both Projection Algorithm" guarantees the correct examining rate, reduces the incorrect captions detection and news story unit segmentation; second, in the condition of situation that perform in a TV program, "Anchorperson Detection Algorithm Based on Background Dominant Color" insures anchorperson detection precision and accurate segmentation of news story unit; third, this subject has explored the video summary generation and it can generate local summary and whole summary; forth, it offers two-level retrieval structure of "key frame - news story unit" and can play segmentation result and the whole video just like a video player."News Story Unit Detection and Retrieval System" , which the arrangement is clearly demarcated, can successfully detect news story unit of special news videos and satisfy the user's real-time retrieval. System goes on a large number of experiments and encloses the result of the test. The later research directions are also pointed out.