Illumination Invariant Data Cost using Modified Census Transform

Stereo matching in non-ideal illumination is a challenging area of research. It assumes identical corresponding color values and this assumption is not guaranteed for real-time environment. As a result most of the stereo algorithms fail to generate good disparity. This paper proposes a Modified Census Correlation (MCC) data cost for stereo matching. The proposed data cost will be derived from modified census transformed indexed image and it is robust to change in lighting direction, exposure and illumination color. The obtained total energy is optimized for disparity estimation using Graph-Cut. An exhaustive evaluation using Middlebury stereo image proves the robustness of the proposed technique for variety of illumination and exposure conditions.