Local adaptive nonlinear diffusion

A non-linear diffusion filtering method of an image with local adjustment of the diffusivity function K factor, is presented. Local image features produced by textures and other local geometric characteristics are detected, where a defined edges proximity in the gradient direction occur. In those regions, a new diffusivity function K factor, proportional to the inverse of the edges proximity, is computed, resulting in a stronger diffusion effect. Then, diffusion becomes locally stronger and faster, resulting in the suppression of local small geometric features and textures, while a good spatial accuracy of the nonlocal image edges is preserved. The model results in an improved segmentation of larger areas. Examples demonstrate how effective is the diffusion in images with strong textures regions.