Color Correction Of Intrinsic Images For Stereo Matching

Stereo matching is one of key steps to infer disparity map from stereo images in computer vision.Most of stereo matching algorithms rely on color similarity,and they generally cannot handle color discrepancies effectively.It is necessary to perform the color correction so as to improve the accuracy of stereo matching.In this paper,we present an intrinsic color correction method to alleviate the color discrepancies between a pair of stereo images in stereo matching,so that the color appearance of one image,i.e.,the target image,is consistent with the other image,i.e.,the source image.The intrinsic decomposition can help us to obtain the intrinsic images of the stereo images.If the stereo images have color discrepancy,their intrinsic images also have color discrepancy.However the degree of color discrepancy is different.Color corrections on the intrinsic images will be more feasible and effective than on the stereo images directly.Our method begins by decomposing both the target image and the source image into two intrinsic layers respectively,i.e.,the reflectance layer and shading layer,using the intrinsic decomposition technique.Then,a practical and effective consistent segmentation algorithm,which is applied to the original stereo images,is proposed to establish the region correspondences locally.Next,two local color correction methods,which are based on the region correspondences,are adopted to transform the two intrinsic layers of the target image,respectively,making them as similar as possible with those of the source image.Finally,the two corrected layers of the target image are combined to yield the final corrected image.The experimental results demonstrate that the proposed method not only enhances visual color similarity between the stereo images but also improves the accuracy of their stereo matching.