Perceptually relevant and piecewise linear matching of silhouettes

  • Authors:
  • Xenophon Zabulis;Jon Sporring;Stelios C. Orphanoudakis

  • Affiliations:
  • GRASP Laboratory, Levine Hall L402, University of Pennsylvania, 3330 Walnut Street, Philadelphia, PA 19104-6228, USA;Department of Computer Science, University of Copenhagen, Universitetsparken 1, DK-2100 Copenhagen, Denmark;Institute of Computer Science, FORTH, P.O. Box 1385, GR-71110 Heraklion, Crete, Greece

  • Venue:
  • Pattern Recognition
  • Year:
  • 2005

Quantified Score

Hi-index 0.01

Visualization

Abstract

In this paper, a novel alignment method for silhouettes is proposed. This method is based on the establishment of correspondences between landmarks on their boundaries and, in turn, on the establishment of correspondences of the boundary pieces in between these landmarks. The method yields more correct correspondences than conventional methods that scale the arc-length descriptions of silhouettes to align them. The selection of landmarks is investigated as to the robustness of their localization and their perceptual relevance. Matching of silhouettes is then achieved by quantifying the dissimilarity of a pair of silhouette boundaries, based on a novel dissimilarity metric. The matching procedure is evaluated, based on retrieval experiments, and it is concluded that the precision of the results is higher than that obtained by conventional pointwise comparison methods.