Skip to Main Content
3D model matching has been widely studied in computer vision, graphics and robotics. While there is much success made in the matching of natural objects, most of these approaches consider smooth surfaces and are not suitable for computer aided design (CAD) models because of their complex topology and singular structures. This paper presents a novel spectral approach for the 3D CAD model matching in the framework of manifold learning. The 3D models are treated as undirected graphs. A regularized Laplacian spectrum approach is applied to solve this problem where the regularization term is used to characterize the shape geometries. Spectral distributions of different models are obtained and then compared by their divergence for model retrieval. The proposed approach is tested with models from known 3D CAD database for verification.