Ontology based viso-semantic similarity for image retrieval

In this paper we present COMPASS/web, a distributed image retrieval system that provides novel solutions to the problem of efficient and effective content-based browsing and retrieval over large multimedia databases. The system aims to exploit the synergy between visual and textual information introducing a viso-semantic distance based on a visual and a semantic similarity measure. The visual similarity is computed using low level features describing the image content while the semantic similarity is computed over graphs representing ontologies. Interactivity is achieved using relevance feedback mechanisms to adapt the similarity measures to the current task exploiting successive hints from the user. Preliminary results and future work directions are presented.