On a spatially varied gradient fidelity term in PDE based image denoising

Image denoising algorithms based on non-linear diffusion PDE models, such as Perona-Malik and total variation denoising tend to converge to a piecewise constant image. Although these models have been demonstrated to preserve edges efficiently while smoothing noises, they often lost some small details like textures. In this paper, we propose a noise removal PDE based algorithm combining with a spatially varied gradient fidelity term. The new model can both avoid the trouble of choosing a proper parameter and preserving better textures and other small details while removing the staircase effect in smooth regions. Experimental results on several classical images illustrate the effectiveness of the proposed scheme in image denoising.