AI-STRATA: A User-Centered Model for COntent-Based Description and retrieval of audiovisual sequences

摘要

We first insist on the need for conceptual and knowledge-based audiovisual models in AV and mutimedia information retrieval systems. We then propose several criteria for characteriing audiovisual representation approaches, and present a new approach for modelingand structuring AV documents with annotations interconnected strata. This consists in analyzing AV documents through analysis dimenisions allowing the detection of objects of interest of any type. Annotations are strutured by annotation elements representing both objects of interest and relationships. A knowledge base is used in order to monitor the annotation process. We show how to use annotations to link different strata on the base of explicit or implicit contexts and how AI-Strata can be used to build contextual views of stratum, using both annotation and knowledge levels. We finally show how the mdoel can efficiently support different description tasks such as indexing, searching and browsing audiovisual material.