The nonlinear diffusion of image filtering is from the idea of heat equations. Its key point is to choose a proper diffusion coefficient and control the diffusion direction. In the previous models, the diffusivity depends on the gradients of images, thus it is easily affected by noises. Furthermore, many fine structures such as textures are prone to being taken for noise and then will be removed. In order to overcome these shortcomings, first, in this paper we introduce a novel computational technique for diffusivity by using the dual tree complex wavelet transform. Second, we develop a nonlinear diffusion model for image filtering. Finally, an image diffusion filtering method based on the dual tree complex wavelet transform and wave atoms thresholding is presented, and also compared with the previous methods. Experimental results show that many features of image such as edges and textures can be preserved well after filtering via the proposed algorithm.