A novel approach of generating video scene structure

It is important to organize the unstructured video data properly for video content analysis and application. Existing shot-based video analysis can not get good performance for effective browsing and retrieval because shots are too numerous to handle and can not convey meaningful semantics. The video scene is a meaningful and high-level semantic video unit. In this paper, we present a simple and quick approach of generating video scene structure. We segment the video into shots with an improved twin-comparison method. Shots are grouped into shot clusters using a sliding shots window method, which can greatly reduce computation effort and time. A correlation function between shot clusters is defined to analysis their correlations and to construct the video scene structure. Experimental results show it is an effective method.