On the reliability of correlation based stereo matching

Abstract

Techniques for enhancing the reliability of correlation- based stereo matching are described. A confidence factor is introduced to evaluate the accuracy of a match measure and to isolate the areas where the matching algorithm fails to provide correct results. Contrast and spatial frequency content, which are major factors affecting the matchability of neighborhoods, are incorporated in the match measure as the confidence factor. The resulting match score should be more useful in determining the areas requiring interpolation and the smoothness of the interpolated values in the removal of bad matches