Topological asymptotic analysis provides tools to detect edges and their orientation. The purpose of this article is to show the possibilities of anisotropic topological gradient in image restoration. Previous methods based on the topological gradient used isotropic diffusion to restore images. These methods are improved here by using anisotropic diffusion and differentiating between principal and secondary edges. A texture detector is also used to increase the diffusion outside textured regions. Numerical results are presented, including a comparison with the Non-Local Means method. The algorithms presented here lead to results similar to the Non-Local Means (in terms of quality), and shorter processing times.