Novel variational technique for mixed Poisson-Gaussian noise filtering

Abstract

A variational framework for mixed noise removal is proposed in this article. The considered variational model is based on an energy functional minimization that leads to a nonlinear second-order anisotropic diffusion-based model. This PDE (partial differential equation) model is then solved numerically by applying an approximation algorithm that is constructed using the finite difference-based method. The obtained explicit numerical approximation scheme is able to restore successfully the images corrupted by both Gaussian and Poisson noise, its effectiveness being illustrated by the described mixed denoising experiments.