Joint planar parameterization of segmented parts and cage deformation for dense correspondence

摘要

In this paper, we present a robust and efficient approach for computing a dense registration between two surface meshes. The proposed approach exploits a user-provided sparse set of landmarks, positioned at semantic locations, along with closed paths connecting sequences of landmarks. The approach segments the mesh and then flattens the segmented parts using angle-based flattening and low distortion boundary constraints. It adjusts the segmented parts with a cage deformation to align the interior landmarks. As a last step, our approach extracts the dense registration from the flattened and deformed segmented parts. The approach is capable of handling a wide range of surfaces, and is not limited to genus-zero surfaces. It handles small features, such as fingers and facial attributes, as well as non-isometric pairs and pairs in different poses. The results show that the proposed approach is superior to current state-of-the-art methods.