Cross-media database retrieval system based on TOTEM

Abstract

This paper presents a model CSRM for realizing cross-media semantic retrieval in multimedia digital library. The object semantic is regarded as the core of all semantic information which is mined by maximum likelihood estimate method. Bayesian network instantiation, Markov chain are respectively utilized to recognize dynamic scene semantic and complex event semantic. Semantic templates are constructed to store the mapping relation between media features and semantic concepts so that all media in multimedia digital library could be automatically annotated. Besides, user could retrieve cross-media information with media example set in real time. A prototype system of cross-media retrieval based on TOTEM is constructed to prove the validity of CSRM and flexibility of data management. © Springer-Verlag Berlin Heidelberg 2006.