| Co-Investigator(Kenkyū-buntansha) |
HAMADA Reiko The University of Tokyo, Graduate School of Information Science and Technology, Research Fellow, 大学院・情報理工学系研究科, リサーチフェロー
IDE Ichiro National Institute of Informatics, Software Research Division, Assistant Professor, ソフトウェア研究系, 助手 (10332157)
SASAKI Shuichi The University of Tokyo, Graduate School of Information Science and Technology, Professor, 大学院・情報理工学系研究科, 教授 (50291290)
|
|---|
| Research Abstract |
Reflecting the increasing importance of handling multimedia data, many studies are made on indexing to TV broadcast video. But there are few studies on video indexing which make use of advantage of textbook. For example, in drama, the order of scenes in the video is same as that in the scenario, but in educational programs such as cooking programs, the order of events are different in the video and in the document. So, in our task, we must gather hints from each medium and integrate them effectively. In addition to that, we take advantage of domain specific constraints and knowledge. In our system, first we create the domain specific dictionaries from many documents. By using the dictionaries, the structure of a document is analyzed. On the video side, cut detection and word spotting are done to detect semantic scene boundaries. Keywords (ingredients, seasonings or, verbs) in a document are specific to each video, so they can be used for limited vocabulary word spotting. Finally, considering the structure of the document and the video, we associate them.
|
