Robust region‐wise colour correction method for stereo matching

Significant colour discrepancies between stereo images have severe impacts on stereo matching algorithms, which depend on colour similarity. To address this problem, a region‐wise colour correction method is proposed in this study. First, the source image, which is to be corrected, is segmented into a set of regions using the mean‐shift method. Scale invariant feature transform (SIFT) features are extracted from both the source image and the reference image, and then matched. The matched SIFT pairs are refined by exploiting epipolar geometry constraint. Based on the segmentation result and the refined SIFT matches, the regional correspondences between two stereo images are estimated. Then, each matched region pair is used to compute a local colour correction function. A set of colour weight maps is calculated for these functions. Finally, to alleviate the colour transformation discontinuities along the region boundaries and facilitate a smooth colour correction globally, the corrected colour is obtained by combining the colour correction functions using the colour weighting maps. The authors apply the proposed local colour correction method to stereoscopic images before conducting stereo matching. The results indicate that the proposed algorithm can effectively and robustly alleviate the colour discrepancies, and improve the accuracy of stereo matching.