Query by sketch and relevance feedback for content-based image retrieval over the web

Content-based image retrieval systems are being actively investigated owing to their ability to retrieve images based on the actual visual content rather than by manually associated textual descriptions. This paper considers the issues related to the porting of such systems to the World Wide Web and proposes some ways to solve them. To substantiate our ideas, we propose a web-based image retrieval system that allows the user to express a query as a simple sketch portraying 'what' she/he is looking for. The system relies on a three-layer relevance feedback architecture to progressively refine retrieval results according to the user's preferences. We also emphasize the use of the vector space model for features representation and the cosine distance for similarity ranking. Performances are presented using both well-assessed information retrieval measures and subjective evaluation criteria.