Multiview Normal Field Integration using Graph-Cuts

In proceedings of Central European Seminar on Computer Graphics for Students, Apr. 2012
 

Abstract

While there are many algorithms which address the reconstruction of partial surfaces from single-view normal fields, to the best of our knowledge, there is only one method [Chang et al., 'Multiview Normal Field Integration using Level Sets', CVPR 2007] which focuses on the reconstruction of the full shape of an object from multiple normal fields captured from multiple viewpoints. In this paper, we propose an alternative approach for the integration of such normal fields. We use a similar energy formulation as Chang et al., but replace the employed level-set representation with a graph-cut based reconstruction. Based on the visual hull of the object we estimate the visibility from each viewpoint. Then we project estimated normal fields to the visual hull and the optimal surface is computed by maximizing the flux of the obtained vector field through the surface. The graph-cut approach allows for a fast and globally consistent optimization of the given problem. Finally, we demonstrate the validity of our algorithm on synthetic data sets.

Bilder

Paper herunterladen

Paper herunterladen

Bibtex

@INPROCEEDINGS{Osep-2012-MultiViewNormalFieldIntegration,
     author = {Osep, Aljosa},
      title = {Multiview Normal Field Integration using Graph-Cuts},
  booktitle = {Central European Seminar on Computer Graphics for Students},
       year = {2012},
      month = apr,
   abstract = {While there are many algorithms which address the reconstruction of partial surfaces from
               single-view normal fields, to the best of our knowledge, there is only one method [Chang et al.,
               'Multiview Normal Field Integration using Level Sets', CVPR 2007]  which focuses on the
               reconstruction of the full shape of an object from multiple normal fields captured from multiple
               viewpoints. In this paper, we propose an alternative approach for the integration of such normal
               fields. We use a similar energy formulation as Chang et al., but replace the employed level-set
               representation with a graph-cut based reconstruction. Based on the visual hull of the object we
               estimate the visibility from each viewpoint. Then we project estimated normal fields to the visual
               hull and the optimal surface is computed by maximizing the flux of the obtained vector field through
               the surface. The graph-cut approach allows for a fast and globally consistent optimization of the
               given problem. Finally, we demonstrate the validity of our algorithm on synthetic data sets.}
}