A non-rigid registration algorithm for compatible skeletonization

This paper presents a registration framework for curve skeleton of 3D models. The algorithm matches two skeletons and deforms a source skeleton to the other target skeleton, so that each node aligns with its closest point on the target skeleton in a non-rigid manner. Transformations defined on each individual skeletal node is optimized to minimize a weighted combination of three errors: the Data Error, Smoothness error, and Length-Preserving Error. The registration can apply to a set of non-isomorphic skeletons robustly and generate a compatible deforming sequence, desirable for many geometric processing tasks. We demonstrate effectiveness of our skeleton matching framework on various 3D models.