Structure-based Shape Correspondence

In recent years, three-dimensional scanning technology develops rapidly. Digital geometry processing aroused extensive attention. And shape analysis has also been extensively studied. The main work of the article is shape matching. Shape matching has applications in many fields, such as shape retrieval, manifold parameters and deformation.Shape matching is divided into rigid matching and non-rigid matching. At present, the non-rigid shape matching is a difficult problem even when the space of deformations is limited to e.g. approximate isometries. A common approach to this problem is to find an invariant representation under deformation of the shapes, and perform the matching in the representation domain. And then, several pairwise geometric relations have been used to shape matching, which contains distance and neighborhood relations. These methods are based on the local information of shapes and distance information. So the correspondence ambiguity problem arises when matching two intrinsically symmetrical shapes. That means, for example, the model of’person’, the right hand corresponds to the left hand and the left hand corresponds to the right hand.In order to solve this problem, we propose a directed-angle scheme, it is used to capture global structural information of the model with distance, and then the graph matching problem is generalized to high-order potentials. Comprehensive experiments demonstrate that our method works well.