摘要
Trailer generation is an important part of movie industry. Editing trailers manually is both time consuming and labor intensive. In this paper, we present an automatic scheme to select candidate movie clips for generating movie trailers. The candidates are well classified and ordered, allowing human editors to choose different types of clips systematically. We analyze professional trailers to guide the training process. The system leverages multimedia analysis techniques to generate candidates for movie trailers. Experiments have shown that clips chosen by the system can significantly outperform randomly picked clips.
- 出版日期2013
- 单位上海交通大学