Iterative reweighted total generalized variation based Poisson noise removal model

Total generalized variation (TGV) regularization model is one of the most effective methods for denoising and eliminating staircase effect. However, the TGV regularization model tends to blur edges as the existence of high-order derivative. In order to avoid the staircase effect while alleviating the edge blurring, an iterative reweighted TGV based Poisson noise removal model is presented under the assumption that each pixel of noisy image follows a Poisson distribution. The weight function incorporated in the TGV regularization term is derived from the expectation maximization (EM) algorithm. We design a new iterative weighted primal–dual algorithm, which is an improvement of the classic iterative reweighted algorithm. Numerical experimental results show the better performance of our model in removing noise effectively while preserving edges and eliminating staircase effect.