Still images can be retrieved by similarity searching on global visual features such as color, texture or shape at the pixel level. Content-based retrieval systems then use and combine all the available low-level features whose computing cost can be prohibitive and they rank the images according to how well they match the submitted quaery-by-example. Finally, they return the best few matches to the used in a ranked result list. But, a subset of features could be sufficient enough to answer very quickly while offering an accepatble quality of results. Moreover, the administration of large collections of images accentuates the classical problems of indexing and efficiently querying inforamtion. Our work focuses on the elaboration of fully aut...