Sports Video Cut Detection And Keyframe Extraction

With the development of multimedia technology, computer performance and networktransmission, digital video have won a wide range of production and propagation. There isan urgent need of users for effective video classification, fast video browsing and retrieval.Content-based video processing and retrieval, which committed to using computertechnology for video semantic classification, automatic indexing and retrieval, contentanalysis and storage management, has attracted the attention of many researchers andinstitutions. Video structure analysis, which is a prerequisite for video content analysis,includes shot segmentation, keyframe extraction and scene clustering. Therefore, shotsegmentation and keyframe extraction is the foundation of content-based video analysis.How to detect shot boundary accurately, extract keyframes effectively and reasonablyhave become an important research area during recent years.After investigating traditional shot boundary detection algorithms, it is implementedthat a video cut detection algorithm based on information theory. The algorithm extractsthe color histograms of the frames, calculates the similarity between two frames usingmutual information. In order to overcome the shortcomings of the traditional methods,frame difference has been calculated by using mutual information between multipleframes, feature vectors has been constructed by frame difference in a siding window, themethod of classification is SVM instead of traditional threshold. After obtaining theboundary information, it is implemented that a keyframe extraction algorithm based onuser attention model. The model contains two parts: motion attention model and staticattention model. The motion attention model focuses on video motion information, whilestatic model concentrates on image information. Then user attention model is beencalculated by fusing motion attention model and static attention model with motionpriority scheme. Finally, keyframes are extracted effectively by user attention model.Experiment results show that cut detection based on information theory has reachedhigh precision rate and recall rate. The keyframe extraction based on user attention modelcan effectively measure the users’ attention, and extract appropriate keyframes from theoriginal video. Our future work will concentrate on video gradual transition detection anda new user attention model by fusing more other features such as audio.