Rapid advances in information technology have led to increased demand for the storage, management, access, and use of multimedia objects. The goal of this research project is to construct a scalable distributed repository for multimedia objects. The high number of dimensions, an important feature of multimedia objects, can significantly increase the search overhead due to increased computations and I/O. One component of the research is to reduce the number of search dimensions using dynamic and incremental transforms. This is supplemented with a performance model that can predict the effect of such dimensionality reduction for different user expectations of recall and precision. Another goal of the research is to decluster the index structure and the associated data on a distributed system. This improves performance as well as availability in case of failures. Building adaptive index structures that can respond to common access and query patterns is also a component of the research. The project is also examining user requirements with respect to consistency of data and supporting different levels of consistency for multimedia objects. The results of the project will be disseminated in the form of scientific papers and software that can be downloaded. The project will lead to efficient index structures and, in general, to better understanding of scalable querying of multimedia data.
