Résumé
This paper introduces a new image representation relying onthe spatial pooling of geometrically consistent visual matches.We therefore introduce a new match kernel based on the in-verse rank of the shared nearest neighbors combined withlocal geometric constraints. To avoid overtting and reduceprocessing costs, the dimensionality of the resulting over-complete representation is further reduced by hierarchicallypooling the raw consistent matches according to their spa-tial position in the training images. The nal image repre-sentation is obtained by concatenating the resulting featurevectors at several resolutions. Learning from these represen-tations using a logistic regression classier is shown to pro-vide excellent ne-grained classication performances out-performing the results reported in the literature on severalclassication tasks.