This research focuses on developing an integrated multimedia information retrieval and database management infrastructure that provides support for multimedia information as first-class objects capable of being stored and retrieved based on their rich internal content. Specifically, techniques are being explored to (1) extract and represent multimedia content at multiple semantic levels useful for retrieval; (2) generalize text-based information retrieval approaches for content-based multimedia retrieval; (3) map high-level queries to lower-level representations supported by the database; (4) perform feature indexing that overcomes the high-dimensionality and non-euclidean nature of multimedia feature spaces; (5) integrate feature indexing mechanisms as access paths in database management systems; and (6) support ranked retrieval in database management systems. Based on the research, a prototype system, entitled Multimedia Analysis and Retrieval System (MARS), that seamlessly integrates multimedia data with structured information traditionally stored in databases is being designed. On the educational front, the existing database curriculum is being modified to incorporate evolving research and development in multimedia information systems. Also, the MARS prototype system is being made available as a teaching, education, and research tool to study human factors issues in supporting content-based retrieval. This research not only has the potential to impact the design of future multimedia information systems but will also prove beneficial to other areas including geographical information retrieval, planning and reasoning applications over database systems. http://www-mars.cs.uiuc.edu
