Shading removal of illustrated documents

Abstract

Pictures of documents have non-uniform illumination causing shading which may yield to bad quality image for human visualization and unsuitable for some image processing algorithms. Most algorithms do not consider the scenario in which documents have large non-uniform regions such as photographs and illustrations. This paper proposes an algorithm to remove the shading of such documents. Once the background is identified, Natural Neighbor Interpolation estimates the shading for non-background pixels. The algorithm performed well on 33 synthetic images using SSIM and PSNR measures. The same quality of performance was confirmed in "real-world" images. © 2013 Springer-Verlag.