Adaptive image retrieval using the fuzzy integral

Abstract

The paper proposes an alternative and iterative image retrieval system which takes into account the subjectivity of human perception of visual content. The proposed system uses a dynamic similarity measure based on the Choquet integral. Both positive and negative user's feedback are modeled by fuzzy sets, and are used to refine the feature relevance weights. The experimental results on more than 3000 texture images illustrate the learning behaviour of the retrieval system.