Résumé: Poisson noise widely exists in photo-limited imaging systems, which is very difficult to remove because of its signal-dependent and multiplicative characteristics. In this paper, we propose a new hybrid regularizer variational model for removing Poisson noise. Based on the weighted hyper-Laplacian prior, the hybrid model combines the overlapping group sparse total variation with the high-order nonconvex total variation (HONTV) as a hybrid regularizer. The proposed model combines the advantages of the HONTV regularizer and the weighted hyper-Laplacian prior with overlapping group sparsity regularizer, it can more effectively preserve sharp edges and details while alleviating the staircase artifacts. To solve the non-convex and non-smooth model, we proposed an efficient alternating minimization method under the framework of alternating direction method of multipliers, where the majorization-minimization algorithm and generalized soft threshold algorithm are adopted to solve the corresponding subproblems. Numerical experiments show that the proposed method has higher quality image recovery than several existing methods.

Langue: Anglais