Fast automated stopping-time and edge-strength estimation for anisotropic diffusion

Anisotropic diffusion (ATD) is an edge-oriented, scale-space based, and iterative image-smoothing process. Two main challenges of ATD are how to automatically stop the iterative process, so to avoid blurring, and how to determine the scale (or edge-strength) parameter, so to best differentiate between edge and noise. In this paper, we propose 1) an automatic noise-adaptive stopping-time estimator and 2) a robust scale parameter (or edge strength) estimator. With these two novel estimators, our adaptive ATD method effectively reduces noise (high PSNR gain) and preserves structures (significantly less blurring) than conventional ATD.