Shape representations and algorithms for three-dimensional model retrieval

• Authors: Thomas Funkhouser;Michael M. Kazhdan
• Venue: Shape representations and algorithms for three-dimensional model retrieval
• Year: 2004

With recent improvements in methods for the acquisition and rendering of 3D models, the need for retrieval of models from large repositories of 3D shapes has gained prominence in the graphics and vision communities. A variety of methods have been proposed that enable the efficient querying of model repositories for a desired 3D shape. Many of these methods use a 3D model as a query and attempt to retrieve models from the database that have a similar shape. In this thesis, we begin by introducing a new shape descriptor that is well suited to the task of 3D model retrieval. The descriptor is designed to enable efficient and meaningful comparison of 3D shapes, thereby satisfying the requirements of efficiency and discriminability that are necessary for an effective, real-time shape retrieval system. We compare our descriptor to other existing descriptors in empirical retrieval experiments, demonstrating that the new shape descriptor provides improved retrieval accuracy and is better suited to the task of shape matching. One of the specific challenges in matching 3D shapes arises from the fact that in many applications, models should be considered to be the same if they differ by a similarity transformation. Thus in order to match two models, a measure of similarity needs to be computed at the optimal translation, scale and rotation. In this thesis, we review a number of approaches for addressing the alignment challenge and provide new methods for addressing this issue that give rise to better shape matching algorithms. Additionally, we present two general methods for improving the performance of many extant 3D model matching algorithms by providing a general framework for augmenting existing shape representations with global shape information characterizing salient shape properties. The first approach leverages symmetry information to augment existing representations with information characterizing a model's self-similarity. The second approach factors the shape matching equation as the disjoint product of anisotropy and geometric comparisons—improving the matching performance of many shape metrics by facilitating the task of shape registration.