A Dual Adaptive Regularization Method to Remove Mixed Gaussian-Poisson Noise

The noise in low photon-counting imaging system can often be described as mixed Gaussian-Poisson noise. Regularization methods are required to replace the ill-posed image denoising problems with an approximate well-posed one. However, the sole constraint in non-adaptive regularization methods is harmful to a good balance between the noise-removing and detail-preserving. Meanwhile, most existing adaptive regularization methods were aimed at unitary noise model and dual adaptive regularization scheme remained scarce. Thus, we propose a dual adaptive regularization method based on local variance to remove the mixed Gaussian-Poisson noise in micro focus X-ray images. Firstly, we raise a new 3-step image segmentation scheme based on local variance. Then, a self-adaptive p-Laplace variation function is used as the regularization operator while the regularization parameter is adaptively obtained via a barrier function. Finally, experimental results demonstrate the superiority of the proposed method in suppressing noise and preserving fine details.