Poisson image denoising based on BLS-GSM method

Abstract

Poisson noise removal is of significant importance for many applications such as spectral imaging, night vision and especially in medical imaging and astronomy. Gaussian scale mixture based methods have been widely used in image denoising. In this paper, we focus on the Poisson noise and propose a new strategy based on Bayesian least squares method for its removal. We begin with a method that removes Poisson noise by reducing it to an additive Gaussian noise with a Variance Stabilizing Transformation. Then we combine the localized version of BLS-GSM method to bring out a new denoising strategy for images corrupted by Poisson noise and experimentally show that it outperforms some of the best existing methods for Poisson noising removal both numerically and visually.