A new anisotropic diffusion-filtering scheme to smooth images with heavy-tailed or binary noise types similar to salt&pepper
noise is presented. The proposed scheme estimates edge gradient from an image that is smoothed or ”regularized” with a median
filter. Its performance was demonstrated on synthetic images that were corrupted by Gaussian, salt&pepper and Weibull noises,
and actual medical images. The visual and quantitative evaluation of the scheme demonstrated comparable or better performance.