Shot Boundary Detection And Classification Of Sports Videos

With the development of the computer science and the network technology,multimedia information retrieval systems have intensive impact on all areas of society.Because traditional methods of indexing and retrieval are based on the pure text,user must use words to describe multimedia information precisely. Therefore,there is a contradiction between the rich content of video information and people's subjective understanding,and it is difficult to describe the video information by several key words.Therefore,the content based video retrieval(CBVR) technology emerges as the times require.The CBVR has two methods.The one is to regard video information as aggregate of absolute frames and images,and make use of image retrieval methods for video indexing and retrieval.The shortcoming of this approach is to ignore the temporal relationship between the video frames.The other is to regard video information as aggregate of several shots,and indexing and retrieval for the video are finished by the shot and key frame.The second method is currently a hot research.The main research issues of CBVR are focusing on video shot segmentation,feature extraction and description,key-frame extraction and video structural analysis.In the dissertation,we focus on the Shot Boundary Detection(SBD) and the Sport Video Classification.Firstly,we summarized the some current algorithms for shot boundary detection.As existing SBD algorithms are sensitive to video object motion and no reliable solution exists to provide accurate shot boundary detection,it still remains an unsolved problem.We propose a new algorithm of shot boundary detection,which employs support vector machine(SVM) as a classifier to detect shot boundary.The proposed SBD algorithm introduces the concept of the visual attention features based on the research results of psychology,which presents advantages in its robustness to video object motion.Extensive experimental results carried out on the TRECVID 2007 database show that the proposed algorithm works well in detecting shot boundary measured by both recall and precision.The other work is the classification of sport videos.We propose a new algorithm of sport video classification based on support vector machine,in which we extract color,texture and motion texture as the useful features.The experimental results show that the algorithm has good performance.