Stereo Matching with Nonlinear Diffusion

  • Authors:
  • Daniel Scharstein;Richard Szeliski

  • Affiliations:
  • Department of Mathematics and Computer Science, Middlebury College, Middlebury, VT 05753. E-mail: schar@middlebury.edu;Microsoft Research, One Microsoft Way, Redmond, WA 98052-6399. E-mail: szeliski@microsoft.com

  • Venue:
  • International Journal of Computer Vision
  • Year:
  • 1998

Quantified Score

Hi-index 0.01

Visualization

Abstract

One of the central problems in stereo matching (andother image registration tasks) is the selection of optimalwindow sizes for comparing image regions. This paper addressesthis problem with some novel algorithms based on iterativelydiffusing support at different disparity hypotheses, and locallycontrolling the amount of diffusion based on the current qualityof the disparity estimate. It also develops a novel Bayesianestimation technique, which significantly outperforms techniquesbased on area-based matching (SSD) and regular diffusion. Weprovide experimental results on both synthetic and real stereoimage pairs.