A Video Summarization Approach Based on Machine Learning

Video summarization is not only the key to effective cataloging and browsing video, but also as an embedded cue to trace video object activities. In this paper, a video summarization approach based on machine learning is developed for automatic video transition prediction. Several novel features are extracted to characterize video boundary, including cut, fade in, fade out and dissolve for facilitating the understanding content structure and domain rules of a video. These features not only can be used to filter negative false alarms caused by illumination changes but also to improve recognition rate of the key-frames. Our approach provides a good view on temporal continuity of video event. Our results have shown that our approach can accurately predict the transitions in a video sequence and would be a practical solution for automatic video segmentation and video summarization.