Video Feature Extraction And Video Footage

With the development of multimedia technology, Internet technology and computer technology, video has already become a very important media type. It can record and store all the visual information of the spacial and temporal domain about the event, so we can get the detail about the content. With more and more usage of video, the size of video data increases consequently. To make better use of video data, we must manage and organize them efficiently, but it's very difficult to manage and make full use of video data, which results in the research of content-based video management. Video shot segmentation, represent frame extraction, feature extraction and normalized description are the foundation of content-based video management, and the main purpose of them is to meet the requirement of content-based video retrieval and browsing. We have performed a research about the three aspects, made a whole study of the virtues and shortcomings of the methods in existence and brought forward new methods with the reference of MPEG-7 multimedia description standard. In this paper we first propose a method to describe the hierarchy structure model of video with the MPEG-7 segment description tools, and present the DDL description. Secondly algorithms for the detection of the video shot boundaries with complex editing and the discrimination of camera motion and gradual transition are presented. Thirdly this paper proposes a motion-based algorithm for represent frame extraction, which can extract represent frames dynamically according to the requirement of the user. It will extract more frames if there is much motion in a shot, and vice versa. The algorithm also takes into account the distribution of the represent frames in a video shot. Finally we have a research about the method to extract MPEG-7 visual descriptors, thereby to perform the feature extraction and normalized description.