![]() ![]() ![]() |
![]() |
|
|
![]() ![]() ![]() ![]() ![]() |
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
Return to Papers Supporting effective and efficient retrieval of multimedia data is a challenging problem in building a digital library. In this paper, we examine the issues related to accommodating multi-modal retrieval of multimedia data (text, image, video and audio), and propose 2M2Net as a generic framework for such versatile retrieval in multimedia digital libraries. The retrieval is conducted based on the integration of multi-modal features including both semantic keywords and media-specific low-level features. This framework is capable of progressive improvement of its retrieval performance, by applying the learning-from-elements strategy to propagate keyword annotations, as well as the query profiling strategy to facilitate effective retrieval using historic information of the previously processed queries. ![]() DiSC'02 © 2003 Association for Computing Machinery |