Comparison of Graph Cuts with Belief Propagation for Stereo, using Identical MRF Parameters

  • Authors:
  • Marshall F. Tappen;William T. Freeman

  • Affiliations:
  • -;-

  • Venue:
  • ICCV '03 Proceedings of the Ninth IEEE International Conference on Computer Vision - Volume 2
  • Year:
  • 2003

Quantified Score

Hi-index 0.06

Visualization

Abstract

Recent stereo algorithms have achieved impressive resultsby modelling the disparity image as a Markov RandomField (MRF). An important component of an MRF-basedapproach is the inference algorithm used to find the mostlikely setting of each node in the MRF. Algorithms havebeen proposed which use Graph Cuts or Belief Propagationfor inference. These stereo algorithms differ in both theinference algorithm used and the formulation of the MRF.It is unknown whether to attribute the responsibility for differencesin performance to the MRF or the inference algorithm.We address this through controlled experiments bycomparing the Belief Propagation algorithm and the GraphCuts algorithm on the same MRF's, which have been createdfor calculating stereo disparities. We find that the labellingsproduced by the two algorithms are comparable.The solutions produced by Graph Cuts have a lower energythan those produced with Belief Propagation, but this doesnot necessarily lead to increased performance relative tothe ground-truth.