Video Summarization based on Film Grammar

Searching time-intrinsic contents, such as movies, drama, is often a time consuming task, since it is not satisfactory to see some of screen shots to catch the whole story. Summarizing such contents is one of the solutions to diminish the cost in browsing and grasp the contents rapidly. Most of the video summarization methods ever proposed extract `conspicuous´ shots from audio/visual point of view, and combine them together to create a summary. These methods imply an issue of disregarding contextual dependency between shots or scenes. We propose a method of summary generation based on `film grammar´, which keeps the dependency between shots or scenes. Experimental results show this method provides a viewer better summaries to comprehend the context