An Alternative Image Retrieval System Based on Visual and Thematic Corpus Organisation

Abstract : Retrieving images in large databases involves either thematic or visual querying capabilities. We think that both approaches can be efficiently combined to provide a thematically and visually relevant retrieval. This paper deals with bi-modal access in image retrieval systems. Our approach relies on a corpus organisation based on both visual and thematical clustering, allowing access to non-indexed images. An adapted retrieval process has been designed to reformulate queries, taking into account users's judgements that deal with thematical and visual aspects of their information needs.