Video Sequence Based Poster Generation

Video summarization is an efficient way for massive video browsing．Current methods are limited in producing an expressive and effective video abstraction．This paper presents a novel method of automatic video poster generation to meet the above two goals．The main issues of our method concern the key frame selection from the video and automatic frame layout on the poster．We introduce a comprehensive visual attention model to extract the desired key frames from the video sequence and sort these key frames by their perception importance．A location importance driven layout algorithm is proposed based on a set of constraints．Experimental results show the effectiveness of our method．