The R2I project aims at conceiving methods for interactive search of images in very large datasets (typically one billion images). It is interesting to note that a recent poll done by IFOP-Omnibus in 2006, studying satisfaction of Internet users, highlighted that 96% of them were on the whole satisfied with the services offered but 75% encountered difficulties in searching information2. What is true when searching information in general is even truer for the case of image search, because of the gap existing between keywords used to formulate queries and semantic information carried by images. In this context, R2I will be oriented towards user needs, by designing and combining several tools making the search of images by content easier. These tools will be able to: • extract semantics from images and transform raw data into a semantically rich representation, • cluster similar images and propose visual summaries of search results, • allow user interaction via semantic concepts related to images, • allow the indexation of very large volumes of images. At the end, this project will come up with a very ambitious system for interactive search, able to index about one billion images and able to provide users with advanced interaction capabilities. The partnership between Exalead, leader company on the market of corporate network indexation and specialist of user centric approaches, Imedia, research group internationally recognized for its work on interactive search of multi-media documents, and Lear, which benefits from an international reputation for his work