A neural network strategy for 3D surface registration

Abstract

3D surface registration is commonly used in shape analysis, surface representation, and medical image aided surgery. This technique is extremely computationally expensive and sometimes will lead to bad result configured with unstructured mass data for its' iterative searching procedure and ill-suited distance function. In this paper, we propose a novel neural network strategy for surface registration. Before that, a typical preprocessing procedure-mesh PCA is used for coordinate direction normalization. The results and comparisons show such neural network method is a promising approach for 3D shape matching. © Springer-Verlag Berlin Heidelberg 2006.