Abstract: The key point of research on anisotropic diffusion is to build the adaptive and stable diffusion model. However, there are still two problems in the existing anisotropic diffusion models --- the determination of the gradient threshold and the iterative stopping time. In this paper, we propose a time-dependent robust anisotropic diffusion method. In the method, the Gaussian scale and the gradient threshold are set to the monotonically decreasing function of the time, which is very useful to accurately extract edge and boundary features. Meanwhile, an iterative SNR measure is defined to effectively determine the stopping time, so as to lessen over-smooth regions. In order to show the validity of the proposed diffusion scheme, we use one gray image ``Pinecone'' in our experiments. Experimental results have shown that the time-dependent robust anisotropic diffusion methods have superiority over the existing methods and they can effectively smooth out noise while preserving edge features.