Anisotropic diffusion based on directional consistency

Abstract

A novel image anisotropic diffusion algorithm is proposed. As noise was more likely to be random, directional gray level consistency gives a feasible discrimination between noise and image structure. Firstly, a measurement of directional consistency was defined. Then, this consistency was transformed into diffusion weight, and diffusivity of different directions was computed from gradient and diffusion weight. At last, anisotropic diffusion was realized based on this directional different diffusivity. Some typical image diffusion algorithms were compared with proposed algorithm. Experimental results on both synthetic and real image with hybrid noise show the efficiency of proposed algorithm.