Multidimensional clustering index for image retrieval

This article addresses the field of content based image retrieval on large collection of images on a query by example basis. We propose a multidimensional organisation of the image database based on several clusterings performed on low level descriptors separately. We show that using this multidimensional index for querying the database outperforms any of the single dimensional query. The other advantage of this method is that it enables to combine several low level descriptors in a single query while keeping the physical meaning of each descriptor in mind. The results on the Amsterdam Library of Object Images are convincing.