A Compressed Video Database Structured for Active Browsing and Search

We describe a unique system called ViBE (video browsing environment) for browsing and searching large databases of video sequences. The system first computes the DC sequence for a given MPEG sequence. It then detects and identifies shot boundaries by using the generalized trace. A hierarchical tree structure is constructed for shot comparison and keyframe extraction. In addition to low-level image features, the system also uses pseudo-semantic features to characterize the frames. Finally, the results are presented to the user in an active browsing environment which we call a similarity pyramid. The users can also prune and reorganize the environment using relevance feedback methods.