Correspondences of Persistent Feature Points on Near-Isometric Surfaces

摘要

We present a full pipeline for finding corresponding points between two surfaces based on conceptually simple and computationally efficient components. Our pipeline begins with robust and stable extraction of feature points from the surfaces. We then find a set of near isometric correspondences between the feature points by solving an optimization problem using established components. The performance is evaluated on a large number of 3D models from the following perspectives: robustness w.r.t. isometric deformation, robustness w.r.t. noise and incomplete surfaces, partial matching, and anisometric deformation.