This paper describes the details of NHK's approach to the rushes summarization task in TRECVID 2008. From a broadcaster's point of view, removing redundancy and attaining a pleasant tempo/rhythm, as well as the recall ratio, are important. Here, we introduce a novel means of representative image selection for comparison which can be applied to remove redundancy. Specifically, we make best use of normal behavior of shooting i.e., being motionless until start trigger, and after the trigger performer and camera move almost the same way (same motion.) We have built and tested a summarization system based on this technique. Although the recall ratio in our evaluation was only 33%, the reduction of duplication and the quality of tempo/rhythm were good. Improvements that can be made to our system are discussed based on the results of this evaluation.
展开▼
