The non-linear diffusion techniques were proposed for overcome the linear diffusion defaults. The linear diffusion was a homogeneous diffusivity with a constant conductivity. In this diffusion process, the noise and the edges were smoothed in the image. In order to prevent the edge from being smoothed during the denoising, the nonlinear diffusion was proposed by Pereona and Malik. In this method, noise was smoothed Simultaneously with the edges blurred. In diffusion processes, the conductivity is dependent on the image local information. We analyzed the ineffectiveness of isotropic and extended the work into the tensor-based anisotropic diffusion. It would be desirable to rotate the flux towards the orientation of interesting features. We compare the difference of isotroic linear and non-linear anisotropic diffusivity, and considere how to design non-linear anisotropic conductivity based on the different requires of the image filtering.