Image indexation and content based search using pre-attentive similarities

In this paper we introduce the general architecture of an image search engine based on pre-attentive similarities. The components of this system are presented and some of them are discussed. Our approach is based on the extraction of interest points using a multiresolution technique. The spatial distribution of these points is embeded in bidimensional histograms of classic features (the first five differential invariants). A user-based relevance feedback is introduced in order to overcome the limitations of any pre-attentive similarities based scheme.