Non-parametric image transforms for sparse disparity maps

In this paper two image transforms are proposed for the calculation of sparse disparity maps. We present a new variation of the Census Transform, which we call the Thesholded Census Transform. This allows the calculation of the pixels around the edges without a separate edge-detection stage. Then we propose a new application of the Complete Rank Transform (which has so far only been used to calculate optical flow) to solve the Stereo-Matching problem. The utilization of both image transforms represents an improvement in error rates and computational cost against the Census Transform, which is the state of the art image transform used for Stereo-Matching.