Spatially varying weighted HSI diffusion for color image denoising

This paper presents a novel anisotropic diffusion framework for color image denoising. Unlike previous approaches, our method is based on separating a color image into hue, saturation and intensity (HSI) components, and then diffusing each component with spatially varying weighted partial differential equations (PDE). The hue denoising is implemented by a new weighted orientation diffusion, and the saturation diffusion is a modified curvature flow. The intensity diffusion PDE is a combination of a gradient vector flow (GVF)-based filter and a fourth-order filter. This combined technique provides a robust and accurate denoising process, i.e., it preserves edges well and at the same time overcomes the staircase effect in smooth regions. The denoising experiments on a set of standard color images have shown satisfactory results.