Multimedia Information Retrieval Systems

点击翻译按钮获取中文摘要

The project designs storage and retrieval methods for multimedia databases, in collaboration with Dr. Christodoulakis, head of the MUSIC-FORTH Center, Chanea, Greece. Specifically, it focuses on high-performance similarity retrieval of multimedia objects by content. For text, it uses signature file methods, inversion and combinations (hybrid methods). For images and other media (voice, etc.), it examines indexing methods for high-dimensionality spaces, treating objects as multi-dimensional points in feature space. Such features could be, e.g., the Fourier-transform coefficients of an image. Parallelism further improves the response time for large objects, using declustering methods, e.g., based on fractals or error correcting codes. The results of this work will be retrieval and declustering methods for several types of media. These methods will serve as building blocks for the multimedia and object- oriented DBMSs of the near future. This project, which is co- sponsored by the European Community Esprit program, will foster collaboration between researchers in the United States and Europe, enhance exchange of research results, and speed up development of a new generation of DBMSs.