Automatic key-frames extraction to represent a video

An efficient video content representation using automatic key-frames extraction is presented in this paper. The proposed video-content representation provides the capability of the more efficient browsing digital video sequences. Firstly, each video sequence is partitioned into shots by applying a shot-cut detection algorithm. We construct multidimensional shot-level feature vector by fusing audio and visual information to describe the average frame properties of the shot. Secondly, shot selection is accomplished by clustering similar shots so that shots of similar content are gathered together. While for a given shot, key frames are extracted based on maximizing Kullback-Leibler divergence criterion for locating a key-frames set. An experimental system has been built up. Experiments verify the effectiveness of the proposed approach.