Anisotropic total generalized variation model for Poisson noise removal

When removing Poisson noise, it is a challenging task to overcome the staircase effect and maintain edge details. To achieve this goal, this paper introduces an anisotropic diffusion tensor into the total generalized variation regularization, and proposes an improved variational model for Poisson noise suppression. The included anisotropic diffusion tensor helps to preserve the structural features of images while denoising. Computationally, we design an efficient alternating minimization method in detail to obtain the optimal solution by combining the classical primal-dual algorithm. Finally, in contrast with several popular regularization models, experimental results show that our denoising model has obvious advantages in staircase reduction and edge preservation. At the same time, our recovered results also have the lowest MSE and the highest PSNR, SSIM values.