Forward-and-Backward Diffusion Based on Orientation Information Measure

Abstract

In this paper, a forward-and-backward (FAB) diffusion model based on orientation information measure is proposed with attempting to address the problems of the classical FAB diffusion model. In the classical FAB diffusion model, the nonlinear diffusion coefficient is locally adjusted and able to switch the diffusion process from a forward to a backward model for adaptive image denoising and enhancement. The nonlinear diffusion coefficient is controlled by the gradient of the image which is undesirable due to the fact that it is sensitive to noise. In the diffusion process, noise may be retained or even amplified because both the edges and noise have large gradients such that they can not be distinguished correctly. In contrast to the classical FAB diffusion model, the proposed model adopts the orientation information measure to control the behavior of the diffusion coefficient. The orientation information measure can detect the edge features correctly and it is not sensitive to noise. Therefore, the proposed model can overcome the difficulties successfully. In its numerical implementation, an efficient numerical schema is used to ensure sufficient numerical accuracy. Experimental results demonstrate that it has better performance than that of the classical FAB diffusion mode and is capable of enhancing the edge features while removing noise.