A stereo image matching method on images with varying light conditions

In the process of 3D regeneration from stereo images, a disparity map can be determined by matching the same objects on the stereo image pair. However, different light conditions of an image pair, such as exposure, illumination, and illuminant color, usually compromise the accuracy of matching. This usually occurs when taking outdoor images. In this paper, we propose a preprocessing method to improve condition of image to build disparity map with less effect from varying light conditions. Stereo images from the Middlebury dataset is used to test our method. We used LogRGB algorithm as a baseline. Compared to the ground truth, experimental results show that our method performs the 3.02% RMS error and 4.0% percentage of bad matching while 3.47% RMS error and 6.9% bad matching for LogRGB.