Interactive key frame selection model

Video summarization can provide a fine representation of the content of video stream and reduce a large amount of data involved in video indexing, browsing, and retrieval. Moreover, Key frame selection is an important step in the research of content-based video analysis and retrieval. Although there exist a variety of methods for key frame selection, they are heuristic and closed systems, which cannot dynamically generate video summary with user’s preference. In this paper, an M-estimator and epipolar line distance constraint camera motion estimation algorithm is introduced as camera parameters is an important motion feature for key frame selection, and Broyden-Fletcher-Goldfarb-Shanno (BFGS) method is applied to optimize estimated parameters. Moreover, since Interactive Computing is a novel-computing model that represents the transition of algorithm to interaction, an interactive model of key frame selection (IKFS) is presented as a result of improving the model of key frame selection (KFS). The model of KFS and IKFS are proved to satisfy the criterion of induction and coinduction, respectively. Experimental results show that the processing scheme generates flexible and desirable summarizations whose distortion rate is lower than current method. Above all, IKFS is an extension to KFS.