Blind document image enhancement based on diffusion process

Abstract

Document image enhancement is a well investigated field of research. Many approaches based on nonlinear anisotropic diffusion have been proved efficient for removing some common artifacts such as bleed-through. However, the proposed methods are somewhat ineffective in removing bleed-through artifacts while strengthening original side strokes. The purpose of this work is to propose a novel nonlinear anisotropic diffusion approach which enhances foreground edges coherence while smoothly removes background artifacts. The proposed method is based on a structure tensor using a new parameter which permits to better distinguish between foreground and artifacts characters. The diffusion scheme is expressed through a new set of equations using the tensor eigenvalues. This system allows to perform a second discrimination between foreground/background strokes. The performance of the proposed enhancement scheme is evaluated by means of objective measures and perceptual judgment of the obtained results on real images.