Stereo correspondence using a fuzzy approach

Abstract

Normalized Cross Correlation (NCC) and Sum of Squared Differences (SSD) are the measures generally used in area-based techniques for stereo correspondence. They fail to establish correspondence in the presence of specular reflection and in occluded regions. Two algorithms for stereo correspondence based on fuzzy relations are presented. A novel idea of finding the correspondence based on Weighted Normalized Cross Correlation (WNCC) is proposed. Experiments with various real stereo images suggest the superiority of these algorithms over Normalized Cross Correlation under non-ideal conditions. © Springer-Verlag Berlin Heidelberg 2002.