Spectral Methods for Isometric Shape Matching and Symmetry Detection
Author: Maksims Ovsjanikovs
Publisher: Stanford University
Published: 2011
Total Pages: 105
ISBN-13:
DOWNLOAD EBOOKShape matching and symmetry detection are among the most basic operations in digital geometry processing with applications ranging from medical imaging to industrial design and inspection. While the majority of prior work has concentrated on rigid or extrinsic matching and symmetry detection, many real objects are non-rigid and can exhibit a variety of poses and deformations. In this thesis, we present several methods for analyzing and matching such deformable shapes. In particular, we restrict our attention to shapes undergoing changes that can be well approximated by intrinsic isometries, i.e. deformations that preserve geodesic distances between all pairs of points. This class of deformations is much richer than rigid motions (extrinsic isometries) and can approximate, for example, articulated motions of humans. At the same time, as we show in this thesis, there exists a rich set of spectral quantities based on the Laplace-Beltrami operator that are invariant to intrinsic isometries, and can be used for both shape matching and symmetry detection. One of the principal observations of this thesis is that in many cases spectral invariants are \emph{complete}, and characterize a given shape up to isometry. This allows us to devise efficient methods for intrinsic symmetry detection, multiscale point similarity and isometric shape matching. Our methods are robust and all come with strong and often surprising theoretical guarantees.
