Research Of Soccer Video Retrieval Based On MPEG

With the rapid development of Digital Video in recent years, such as VOD, Digital TV, Digital Library, Video Conference and Tele-education, it has been accepted and known well by more and more people. Dealing with teeming of vast video data, the text-based retrieval can not satisfy people's need. Since 90's, a new research field called Content-Based Video Analysis and Retrieval (CBVAR) comes into existence. According to the support of analysis and comprehend from low-lever to high lever for the video data, it gets the content of video data and does retrieval based on the content. It can do some efficient inquiries, index, browsing, search and retrieval based on video information, and this work is directly launches based on the content of video data.Retrieval of sports video (TV relay of sports match) as an important branch of CBVAR has always been paid more attention. Comparing with other video, sports video has its own characteristic. First, there are some semantic information of its own field in sports video, such as field goal in soccer and basketball. This semantic events is always the most valuable part in the whole video, so it is has important significance to do retrieval of the semantic events. As one kind of sports video, soccer video has a lot of audience, in order to satisfy the need for different people, it is necessary to do soccer video retrieval based on semantic information. So, we propose an algorithm of shot detection and highlights retrieval for soccer video.In this paper, firstly, we introduce structure of CBVAR, in which the algorithm of shot detection and the retrieval of key-frame are detailed introduction. The task of shot detection is divides video data into physical units-shot. Shot is a group of frame sequence which is uninterrupted shoot by camera, and it is often regarded as the least unit of video. Shot detection can be classified into compressed domain (MPEG) and uncompressed domain with usual algorithm such as based on comparison of histogram differences, pixel differences, clustering, based on DCT coefficient and based on MB. Key-frame is a description of frames in a shot, and it reflected the main content of a shot. Comparing with original video data, it is more efficient to use key-frame to be an index. The algorithms of key-frame retrieval are based on shot, based on average, based on content analysis and based on clustering.Because of the compression of test video in this paper is MPEG, MPEG compression standard is analyzed in this paper, which contains structure of MPEG, encode type and decode type.According to the analysis of the difficulties in soccer video retrieval, we propose an algorithm of soccer video shot detection based on DC image. Firstly, we choose the histogram of Y channel in YUV color space and the dominant-color ratio in frames as the characteristics. Secondly, we get DC image sequence in the MPEG compressed soccer video as the original data for the step-variable shot detection, because of two distinct advantages of DC image: one is DC image contains most information of the frame; another is the store of DC image Obviously reduced and processed easily and quickly. Then we use dominant-color ratio to detect the visibility of field. If it is high dominant-color ratio, we use large threshold, else we use small threshold. Finally, considering there are many cut and gradual shots in soccer video, traditional algorithm can not detect shot exactly, so we propose a new algorithm called step-variable to detect cut and gradual shots.Retrieval of key-frame is very important for the highlights retrieval. For soccer video, one key-frame is not enough to express the content of a shot. Thus we classify the shots into two kind, one is Field-See shot, another is Field-not-see shot. For the Field-not-see shot, because of its short time and less of highlights, we directly get the first frame and last frame the key-frame. For the Field-See shot, because of its small difference between frames but most highlights, we get four frames to be the key-frame based on the same time distance.It is more difficult to retrieval the highlights directly from soccer video, because it only occasionally appears, the position and time of highlights can not be confirmed beforehand. But according to the analysis of highlights, we find that there is a time-relationship between shots. For example, in the highlights of goal, there are three kinds of shots: the shoot for the ball moving towards goal, the celebration of players, and slow motion of the goal. These shots are time-continuous in the soccer video, so we can make it the key for goal retrieval. In this paper, we propose an algorithm of soccer highlights retrieval based on the attribute of shot. According to the shot detection of the correlative shots in highlights, we can realize highlight retrieval from soccer video.Finally, by the operating system of windows xp and software of visual C++ 6.0, we implement a shot detection and highlights retrieval system for soccer video. According to the analysis of the experiment result, the algorithm proposed in this paper has a good effect for soccer video.