Structure preserving inhomogeneous diffusion in image filtering

Abstract

We introduce a new theory relating the magnitude of the image surface normal to an inhomogeneous diffusion that solely diffuses (averages) the mean curvature of the image surface. We discuss the remarkable properties of this diffusion stressing the regularity it imposes on regions and boundaries while preserving the locality of edges and lines. Experiments demonstrating the excellent performance of the algorithms in the areas of noise removal and enhancement are presented.<>