Anisotropic diffusion can remove noise to some extent in image processing. However the contradiction between diffusion and preservation still exists. In this paper, a new nonlinear diffusion model for image noise removal and feature preservation is presented. This model treats inhomogeneity region and image feature adaptively by discontinuity measure and local gradient information. A well balance between diffusion and preservation is also made in this new diffusion method. Experiments results show that the proposed method has high performance compared to other literature methods and is an ideal edge-preserving filtering method. In addition, we use block-based noise estimation to estimate deviation in diffusion equation.