A new automatic framework for document image enhancement process based on anisotropic diffusion

In the last two decades many nonlinear anisotropic diffusion-based approaches have been proposed to deal with document image enhancement. However, all these methods are based on an iterative process that highly depends on two crucial parameters K±, used to separate coefficients representing foreground edges from those representing artifacts into eigenvalues matrices λ±. These parameters are tuned manually. In this paper, a new approach which blindly and automatically highlights eigenvalues coefficients representing foreground strokes is proposed. This solution is then integrated into a full anisotropic diffusion-based filter proposed earlier. The performance of the proposed method is evaluated and compared with non-automatic methods of the state-of-the-art by means of objective measures and perceptual judgment on DIBCO databases and some document images collected from the web.