Prosemantic image retrieval

Abstract

In this technical demonstration we present a content-based image retrieval system based on the 'query by example' paradigm. The system effectiveness will be proved for both category and target search on two standard image databases, even without a "good" initial example and ancillary information, such as device metadata, text annotations, etc. These results are obtained by incorporating in the system our recently proposed prosemantic features coupled with a relevance feedback mechanism, and by maximizing novelty and diversity in the result sets. © 2012 Springer-Verlag.