Correspondences between Salient Points on 3D Shapes

Universität Bonn, Technical Report number CG-2006-3, Juni 2006
 

Abstract

Establishing correspondences between salient points on 3D shapes results in a function mapping similar parts of 3D objects. In this paper, we present a new method for establishing binary correspondences between salient points on 3D shapes. Our algorithm is independent of the shape representation and the object topology. It does not require any prepositioning of the objects. Our method first detects stable salient points that are representative for certain parts of the shape. For each of these salient points it then computes an associated local shape descriptor. We introduce a matching energy between the salient points in two shapes which depends on the similarity of the descriptors and the spatial relationship of the salient points. An iterative optimization scheme determines a correspondence mapping between the salient points which minimizes this energy. The resulting binary correspondences between the salient points can be used for applications like 3D shape retrieval based on similarity estimation, classification of 3D objects, editing, and statistical shape analysis. It is especially useful as an initialization method for approaches relying on prior knowledge about corresponding points like cross parameterization or morphing.

Bilder

Paper herunterladen

Paper herunterladen

Bibtex

@TECHREPORT{cg-2006-3,
       author = {Wessel, Raoul and Novotni, Marcin and Klein, Reinhard},
        title = {Correspondences between Salient Points on 3D Shapes},
       number = {CG-2006-3},
         year = {2006},
        month = jun,
  institution = {Universit{\"a}t Bonn},
     abstract = {Establishing correspondences between salient points on 3D shapes results in a function mapping
                 similar parts of 3D objects. In this paper, we present a new method for establishing binary
                 correspondences between salient points on 3D shapes. Our algorithm is independent of the shape
                 representation and the object topology. It does not require any prepositioning of the objects. Our
                 method first detects stable salient points that are representative for certain parts of the shape.
                 For each of these salient points it then computes an associated local shape descriptor. We introduce
                 a matching energy between the salient points in two shapes which depends on the similarity of the
                 descriptors and the spatial relationship of the salient points. An iterative optimization scheme
                 determines a correspondence mapping between the salient points which minimizes this energy. The
                 resulting binary correspondences between the salient points can be used for applications like 3D
                 shape retrieval based on similarity estimation, classification of 3D objects, editing, and
                 statistical shape analysis. It is especially useful as an initialization method for approaches
                 relying on prior knowledge about corresponding points like cross parameterization or morphing.},
         issn = {1610-8892}
}