A fast non-local based stereo matching algorithm using graph cuts

Abstract

In this paper, a new segment-based stereo matching algorithm (FastNL_Gc) is introduced. FastNL_Gc depends locally on the introduced Non-Local-Mean (NL-Mean) approach that provides a fast denoising technique with an edge preserving property in the initial disparity map estimation and globally on graph cuts for the disparity plane assignment using a new energy formulation of the stereo problem in segment domain. The methodology is tested on Middlebury stereo benchmark and the results indicate that the proposed FastNL_Gc method is compatible with the current state-of-the-art stereo matching algorithms in dealing with the conventionally difficult areas, such as textureless regions, disparity discontinuous boundaries and occluded portions.