Research On Video Concentration Technology Base On Clustering Algorithm

Today, with the widespread use of Internet technology development and video image processing technology, diversity of digital video have emerged in all areas, we can more convenient and faster access and dissemination of video files. But for the emergence of numerous video files, it is an urgent need to resolve the issue that how to quickly browse the contents in order to find the information we need. In this context, video concentration technology has emerged to allowing users quickly browse the contents of the entire video, which is extract critical information from the original video through key frame extraction.Although the video concentration technology has been developed for an extended period, but many problems still exist. First, most current video concentrate algorithms is inefficient, there is too much redundant frames in the extracted thumbnail video. Second, we can't extract extremely effective thumbnail content from the video with complex scenes, while the conventional methods will result in misjudgment. So in order to solve these two problems, it is essential to ameliorate and enhance the efficiency of the video concentration technology.This paper summarizes and analyzes some key technologies such as scene detection algorithm, color space model, key frame extraction algorithm used in the video concentration process. Video concentration system needs to deal with videos that contain complex scenes, if using a single image feature may cause false detection of scene boundaries and image similarity inaccurate measured. This paper is a work of combining multiple image features to perform video concentration. Using histogram calculation color feature, wavelet transform is employed to extract texture feature, fusion of multiple features is able to accurately calculate the similarity between images. For many of the current key frame extraction algorithm cannot solve the complex scenes, I proposed a key frame extraction algorithm based on cluster analysis, and improved the original clustering algorithm. Then completed the video concentration system that can extract an abbreviated video generalize the content information of the original video. Finally, comparison between the experimental results of different methods proved research and technique proposed in this paper has a relatively better effectiveness.