Nonlinear diffusion filter based on Partial Differential Equation (PDE) is good at removing Gauss noises, but do not well at pulse noises. From the diffusion coefficient of Perona & Malik model, the function nature was analyzed. By changing the form of the threshold value in diffusion function, it can get rid of the noises from image while preserving edges better, and overcome the contradictions between keeping edge and eliminating noises.
