This paper presents a browsing and similarity search method of videos based on cluster extraction from graphs. Videos are segmented into shots and represented as the sequence of symbols. The directed graph of videos is constructed from the relationship between those symbols. Initial and terminal shots are extracted from it, and they are displayed for users sequentially from the initial shots to the terminal ones. The shots selected by means of this browsing are used as a query video on similarity search. The performance of the proposed method is examined by using the video dataset of NASA.
展开▼