A mixed noise removal algorithm based on multi-fidelity modeling with nonsmooth and nonconvex regularization

In this article, we propose a mixed-noise removal model which incorporates with a nonsmooth and nonconvex regularizer. To solve this model, a multistage convex relaxation method is used to deal with the optimization problem due to the nonconvex regularizer. Besides, we adopt the number of iteration steps as the termination condition of the proposed algorithm and select the optimal parameters for the model by a genetic algorithm. Several experiments on classic images with different level noises indicate that the robustness, running time, ISNR (Improvement in Signalto-Noise ratio) and PSNR (Peak Signal to Noise Ratio) of our model are better than those of other three models, and the proposed model can retain the local information of the image to obtain the optimal quantitative metrics and visual quality of the restored images.