The Research Of Key Frame Extraction Algorithm In Video Retrieval Technology

With the rapid development of multimedia and Internet technology, video as one of the main form accessing to information resources, has brought us the visual and auditory enjoyment. But it is not easy to find the valid information quickly from the vast amounts of video. Because the video is different from the other data, its structure is complex; it has a large quantity of data, and rich content. The video retrieval method based on video content is to analysis and understand the multimedia data content through computer, extract the color, shape and texture of video visual features as index, in order to query for users, to ensure that efficient multimedia content can be quickly retrieved.Retrieval based on content is usually detected the shot of video first, the video divided into several shots. The image frames in each shot are contained a lot of redundant information. Therefore, this problem is solved through extracting the key frames that can represent the shot content from each shot. The key frame sequence is to represent the video content changes, in order to replace the corresponding video. In that way, it not only could reduce the data need to process, but also improve the efficiency of the retrieval greatly. Key frame extraction is an important method that transformed the video into images. There are many key frame extraction algorithms at present, for example the methods based on shot boundary, motion analysis or clustering. These methods are need to segment the video into shots, and then analyse the color, shape, texture and other features of the image frames, as well as the mutual relations between frames, to find key frames representing the video information. But the key frames extracted sometimes are not necessarily representative of the effective video content, like this the next video retrieval processing has got preliminary error. Therefore, how can the key frames represent the main video content in maximum, and can be applied in different video. This problem has important academic significance and practical value.Video feature extraction is the first step of the key frame extraction; the effect that the key frames describe the shots is directly influenced by feature vectors. In this thesis, the characteristics of video information are introduced from two levels of video and image. And the extraction methods of color, shape, texture and movement feature are explained in detail, quantitative description of these characteristics through the mathematical formulas.Clustering method is effectively to eliminate the correlation of the shots, but it cannot preserve the time order and dynamic information of the original shots. Visual content analysis method is insufficiency to describe the video content for the shots with more motion. In order to take into account the timing and adaptability, a new key frame adaptive extraction algorithm based on sub shots is proposed in this paper. The sub shot segmentation function is suggested utilizing the sample distance separability criteria. The shots are segmented into several sub shots through looking for the method looking for the maximum of the function. This method not only can maintain the video timing, but also avoid the shortcoming that the threshold value segmentation method is not accurate. And then the number of key frames of each sub shot is adaptively determined according to the content changes of sub shots. In this way, the deficiency of existing algorithms that the key frame number extracted is not ideal is solved.The video shots are analyzed from two levels of sub shot and image frame first in this method. Sub shots are detected from shots, and key frames are extracted adaptively. So the users can quickly understand the video according to small amount of visual data.The key frames can be extracted with a higher accuracy rate according to the complexity of the video itself. The performance of the improved method is to measure through fidelity measure and compression ratio. And a new evaluation index of shot reconstruction is defined; the method of this paper is compared with the traditional algorithms. The experimental results show that the improved method of key frame extraction can effectively express the main content of shots, maintain the shot timing and has good robustness.