Top Papers Projects Teaching Location Various

Content Based Video Retrieval

Motivation

Digital video over the Internet is the next "big" medium. The resulting accumulation of video content brings problems in archival and foremost in retrieval of digital video, particularly the formulation and processing of queries. Ideally, these processes can be automated e.g., by analyzing a video's content rather than relying on textual annotations. Unfortunately, automatic indexing and feature extraction from digital video is even harder than still-image analysis. Presently, automatic analysis of digital video is mostly restricted to simple content based features. We developed a framework suitable to immediately explore the consequences of content-based video retrieval with a high granularity of video content.

Synopsis

Our framework employs semantic networks to represent video contents on a high level of abstraction and uses time-varying sensitive regions to link objects in a video to the knowledge base. A prototype was implemented under NeXTSTEP, exploiting the rich user-interface capabilities of this platform to feature drag & drop queries and authoring of the video retrieval system.

Queries are processed by flooding activation from activated nodes in the network to neighboring nodes connected either through associative or inheritance relations. Nodes with a graphical representation can be activated by dragging their visible portion onto a query processing field. Nodes without a graphical representation (e.g., nodes representing actions) can be managed and used as a basis for queries through the node browser.

Selected Publications