Für eine korrekte Darstellung dieser Seite benötigen Sie einen XHTML-standardkonformen Browser, der die Darstellung von CSS-Dateien zulässt.


Video Retrieval Using Artificial Intelligence Techniques

Dr. Kimiaki Shirahama, University of Kobe

Date: 08/04/2011
Time: 14:15-15:45
Room: H-F 114

I present four methods which utilize artificial intelligence techniques to efficiently retrieve interesting videos from a large amount of videos. The first method adopts query-by-keyword approach where sequential pattern mining is used to extract patterns. Each pattern represents features in adjacent shots associated with a certain keyword. However, due to the lexical ambiguity, all of queries cannot be appropriately described by keywords. Thus, the second method adopts query-by-example approach where a query is represented by providing example shots. Particularly, to retrieve a variety of shots taken by different camera techniques and settings, rough set theory is used to extract multiple classification rules which identify different subsets of example shots. The third method aims to improve the retrieval performance using a video ontology as knowledge base. This represents object properties and relations to filter irrelevant shots to a query based on object recognition results. The last method targets scene-level retrieval where annotations assigned to fragmentary shots are organized into a scene. Time series segmentation is used to extract scenes which are characterized by certain patterns of character appearance and disappearance. Through the four methods, I discuss the effectiveness of artificial intelligence techniques for extracting high-level information from lower-level features/annotations in videos.