Optimization of Sparse Color Correspondences for Color Mapping

摘要

This article addresses the problem of finding correspondingcolors between multiple views of a same scene in order tocompensate color differences by color mapping. Both, the denseand the sparse feature matching are studied in the literature toachieve those corresponding colors. However both methods sufferfrom spatial precision and occlusion. Moreover, in case of sparsefeature matching, the spatial and the color space coverage arelow. Therefore, it is difficult to generalize for colors where directcolor correspondences are not known. Though dense featurematching may address this problem, it needs computational effortand may introduce additional occlusion errors. Therefore, in thiswork, we propose to consider the spatial neighborhood aroundsparse feature matching to select the “stable” correspondingcolors. We estimate a color mapping model from the colorcorrespondences which is able to compensate the color differencesbetween the views. We compared the quality of several colormapping methods in a performance evaluation framework. Fromexperimental results, we found that consideration of neighborhoodcan significantly increase the precision of color mapping in spiteof increasing uncertainty of correspondence. Benchmark testsshow good performance compared to recent methods from theliterature.