The domain of image retrieval is still an open problem in comparison with the results obtained in the domain of text retrieval. The problem of semantic image retrieval is even more difficult and open. In this article, we propose a new technique of image retrieval enabling, by interaction with the user, to extract fast and easily his/her objective. Here we propose a 2D graphic interface adapted to the problem of image retrieval and enabling a bidirectional communication: from the system towards the user to visualize the current research results and from the user towards the system so that the user can provide some relevance feedback information to refine his/her query. Relevance feedback is only in the form of validation of result images (or positive examples). This approach has been implemented and tested on an image database of 1000 images. The results are promising. A comparison with a traditional interface is shown.