Removing illumination from image pair for stereo matching

Stereo vision has long been studied and a great variety of algorithms have been proposed to estimate the disparities from the stereo image pairs. However, when stereo matching algorithms are applied to real world images, especially out door image sequences under sunlight, they can hardly produce desirable disparity maps as they perform on standard stereo image pairs. Consequently, we proposed a method based on histogram equalization and tophat filter to mitigate the influence brought by illumination. Then, belief propagation is used in our method to generate the disparity map. In order to evaluate the effectivity of the proposed methods, some experiments are executed with two outdoor stereo sequences and the results indicate that disparity maps after preprocessing through our method perform much better than directly using stereo matching algorithms.