Nonlinear filtering of MR images using geometrically and statistically controlled diffusion

In this paper a novel approach to the filtering of multivalued Magnetic Resonance (MR) images is proposed. The proposed method is essentially a nonlinear diffusion with a statistically and geometrically controlled conductance. The user is required to define samples of individual tissue classes in the input image, and their statistics are exploited during the image filtering. The method can be used in medical diagnostics for the enhancement and segmentation of medical images.