Local stereo matching under radiometric variations

Stereo matching is an active research area in computer vision for decades. Most of the existing stereo matching algorithms assume that the corresponding pixels have the same intensity or color in both images. But in real world situations, image color values are often affected by various radiometric factors such as exposure and lighting variations. This paper introduces a robust stereo matching algorithm for images captured under varying radiometric conditions. In this paper, histogram equalization and binary singleton expansion are performed as preprocessing step for local stereo matching. For the purpose of eliminating the discrepancy of illumination between reference image and corresponding image in stereo pair, the histogram equalization is first explored to remove the global discrepancy. As the second step, binary singleton expansion is performed to reduce noise and normalize histogram results for window cost computation efficient. Afterwards, local pixel matching on preprocessed stereo images is performed with Sum of Absolute Difference (SAD) on intensity and gradient. Finally, the final disparity map is obtained by left-right consistency checking and filtering with mean shift segments. Experimental results show that the proposed algorithm can reduce illumination differences and improve the matching accuracy of stereo image pairs effectively.