A framework for 3D CAD models retrieval from 2D images

The management of big databases of three-dimensional models (used in CAD applications, visualisation, games, etc.) is a very important domain. The ability to characterise and easily retrieve models is a key issue for the designers and the final users. In this frame, two main approaches exist: search by example of a three-dimensional model, and search by a 2D view or photos. In this paper, we present a novel framework for the characterisation of a 3D model by a set of views (called characteristic views), and an indexing process of these models with a Bayesian probabilistic approach using the characteristic views. The framework is independent from the descriptor used for the indexing. We illustrate our results using different descriptors on a collection of three-dimensional models supplied by the car manufacturer Renault. We also present our results on retrieval of 3D models from a single photo. MOTS-CLÉS : modèles 3D, vues caractéristiques, indexation, photos.