Efficient shape retrieval by parts

Modern visual information retrieval systems support retrieval by directly addressing image visual features such as color, texture, shape and spatial relationships. However, combining useful representations and similarity models with effcient index structures is a problem that has been largely underestimated. This problem is particularly challenging in the case of retrieval by shape similarity. In this paper we discuss retrieval by shape similarity, using local features and metric indexing. Shape is partitioned into tokens in correspondence with its protrusions, and each token is modeled by a set of perceptually salient attributes. Two distinct distance functions are used to model token similarity and shape similarity. Shape indexing is obtained by arranging tokens into a M-tree index structure. Examples from a prototype system are expounded with considerations about the effectiveness of the approach.