Video Summarization Based on Semantic Feature Analysis and User Preference

A personalized video summarization system is proposed in this paper. "Personalized" means that each video can be summarized according to viewer\´s own preference. To meet user\´s preference, semantic features of each frame are detected so that their relevance to user preference can be determined. Users can be also capable of setting time or frame number constraint for video summary via a friendly interface. To summarize a video efficiently and effectively, a constrained optimization problem (subject to time constraint and video smoothness) is faced and solved to determine the non-uniform sampling rates for shots relevant to user preferences. Subjective tests show that the quality of the summarized video has a MOS of about 4.0, and the comprehension about the video contents is good.