X Zhou and P Boulanger (2011). Illumination Invariant Stereo Matching Based on Normalized Mutual Information and Census Methods. In Computer Graphics International 2011, 4 pages, Ottawa. June 2011. University of Ottawa.

Stereo matching aims at finding corresponding pixels from two or more images where distance information is computed by triangulation. Due to different illuminations, intensities are not reliable to be used to search for corresponding pixels. In this paper, we propose a novel local matching method which is capable of dealing with illumination variations between cameras. This new method can distinguish pixels in the same window but with different disparities allowing for larger window to be used. Moreover, a more precise matching cost function is used to find the correspondence. The proposed method is compared with five other local stereo methods and is proven to be more robust and effective under various illumination conditions.

Back to to publications.