Research On Technologies Of Hybrid Features Based Video Structuring And Summarization

With the advance of technologies of electronic and information processing, video is becoming more and more important in society today. However it is still difficult to browse, edit and retrieval video data because of its linear and non-structuralized format. Video structure analyses and summarization have been brought up to solve such problems. Video structure analysis is to get back the shots and scenes that disappeared in the edition stage of the video program, while video summarization is to get some representative frames or video segments. Users can browse, edit and retrieval video data freely through a framework which was prepared through video structure analysis and summarization.Hybrid features based video structure analysis and summarization use the features both from the upper level and lower level to strength the description of shot key frames, in order to boost the precision of video structure analysis and representative of video summarization. First, color and edge direction is used to detect shots from video data quickly. Second, bag-of-word model is used to analysis semantic feature of shot key frames according to a semantic model, then upper level features is formed. Third, tow pass window sliding is used to detect the boundaries of scenes using upper and lower level features. Finally, when finished detection of the key shots of all scenes based on similarity of shots and their length and so on, chapter video summarization is generated according to length of scenes.Experiments suggest that hybrid features based scene detection boosts the recall and precision rates compared with traditional lower features based ways. The scene key shots and video summarization have met the demand of navigation system of video desk, and are convenient for user to browse and edit the video data.