In View Of The News Video Content Analysis

Nowadays, with the development of multimedia technology and network, videos are used widely in many kinds of domains. But there are so many videos which have no information of their structures. How to organize and retrieve them well are the urgent needs in the field. As a result, content based retrieval technology is proposed.The research objectives of this thesis are the techniques used in content-based retrieval. We take the video structure analysis as the masterstroke, and then based on the research work we have done, we focus on the introduction of several key techniques, i.e. shot segmentation (or shot boundary detection), scene segmentation (or scene boundary detection), as well as key frame extraction.In this paper, we focus on news video, so algorithms based on news video are proposed as key techniques. We proposed a shot segmentation algorithm which uses adaptive canny descriptor to detect religions containing words. This method also used edge feature, and can remove the interference of subtitles. It turns out to have nice effect in our experiment. Referred to scene segmentation, our method uses the feature that shots containing reporters always have similar contents. We use cluster algorithm, upper body detection, post process to filter the bad shots and find out the best ones. This algorithm can detect in high speed and precision, no matter the dressing background and people changes, thus has a great of practical significance.