Fast Cost Computation Using Binary Information for Illumination Invariant Stereo Matching

Stereo matching methods are used to estimate disparity values from captured stereo images. They exploit characteristics of binocular disparity for disparity estimation. Stereo images captured under practical conditions have different illumination status, which causes disparity errors in the matching operation. To solve this problem, we can use the adaptive normalized cross correlation (ANCC) as a similarity measure that is independent from illumination factors and provides good matching results under various radiometric conditions. However, it has a very high computation complexity because of the bilateral filtered block matching operation. In this paper, we propose a new stereo matching method using binary information to reduce the computation complexity of ANCC. The proposed method uses a global mean value, instead of the bilateral filter. A census transformation is also applied in cost computation for fast block matching.