Fast 3D shape retrieval method for classified databases

In this work, we present a new technique to speed up 3D shape retrieval. Instead of performing an exhaustive search over the whole database, which implies a systematic comparison of the query object with all 3D objects in the database, we restrict the pattern matching to a subset of “good candidates” (the most similar to the query). Assuming that the database has been partitioned into several classes, our retrieval algorithm focuses on the class containing objects that are similar to that of the query. Thus, the systematic pattern matching is performed within the selected class only. The key idea is to index each class by a suitable referent object that will be used to seek the right class on a search request with a given 3D object. In this paper, we describe how to choose the best representative within a class, and then couple it with a 3D retrieval method proposed in the literature. We illustrate the efficiency of our approach trough some experiments.