Analysis of interior point methods for edge–preserving removal of Poisson noise

The aim of this paper is to analyze the behavior of the interior point (IP) approach for an image processing application that requires to solve a large scale nonlinear programming problem, such as the denoising of an image corrupted by Poisson noise. We devise two dierent IP algorithms following the two well known globalization strategies, line search and trust region, with the aim to obtain an acceptable compromise between convergence rate and computational cost per iteration. We show that the obtained algorithms can be useful for computing high accuracy solutions.