The selection of regularization parameter for anisotropic diffusion in image processing is discussed.An approach of noise estimation based piecewise constant image model is proposed.In this approach,image noise is estimated by minimal variance of region in image.And then the relation between regularization parameter and image noise is analyzed.A modified formula between regularization parameter and image noise is suggested.According to this formula,regularization parameter is adapted to image noise.Finally,the rule to select size of gauss mask to regularization parameter is given.Resutt shows that anisotropic diffusion have good adaptability to image noise with the selection of regularization parameter.