Intrinsic shape matching via tensor-based optimization.

Abstract This paper presents a simple yet efficient framework for finding a set of sparse correspondences between two non-rigid shapes using a tensor-based optimization technique. To make the matching consistent, we propose to use third-order potentials to define the similarity tensor measure between triplets of feature points. Given two non-rigid 3D models, we first extract two sets of feature points residing in shape extremities, and then build the similarity tensor as a combination of the geodesic-based and prior-based similarities. The hyper-graph matching problem is formulated as the maximization of an objective function over all possible permutations of points, and it is solved by a tensor power iteration technique, which involves row/column normalization. Finally, a consistent set of discrete correspondences is automatically obtained. Various experimental results have demonstrated the superiority of our proposed method, compared with several state-of-the-art methods. Highlights • We propose a non-rigid shape correspondence algorithm. • We incorporate both the local and global potentials into the similarity measure. • Our method is fully automatic and easy to implement. • It generates quality correspondences for a wide variety of model pairs. [ABSTRACT FROM AUTHOR]