Mean curvature evolution and surface area scaling in image filtering

Proposes a geometrical surface representation for an image, and introduces a nonlinear adaptive filter that diffuses the surface in time at a speed proportional to the mean curvature. Linear variations in intensity (edges) are inclined planes of vanishing mean curvature, and are thus invariant. Noise is characterized by high mean curvature and will be diffused. The authors show that this diffusion resolves the conflict of removing noise while preserving edges. A novel nonlinear scale space filtering relating surface area to the diffusion speed is introduced resulting in very efficient algorithms. Experiments demonstrating excellent performance and efficiency are presented