Abstract:Plug-and-Play (PnP) methods are efficient iterative algorithms for solving ill-posed image inverse problems. PnP methods are obtained by using deep Gaussian denoisers instead of the proximal operator or the gradient-descent step within proximal algorithms. Current PnP schemes rely on data-fidelity terms that have either Lipschitz gradients or closed-form proximal operators, which is not applicable to Poisson inverse problems. Based on the observation that the Gaussian noise is not the adequate noise model in this setting, we propose to generalize PnP using theBregman Proximal Gradient (BPG) method. BPG replaces the Euclidean distance with a Bregman divergence that can better capture the smoothness properties of the problem. We introduce the Bregman Score Denoiser specifically parametrized and trained for the new Bregman geometry and prove that it corresponds to the proximal operator of a nonconvex potential. We propose two PnP algorithms based on the Bregman Score Denoiser for solving Poisson inverse problems. Extending the convergence results of BPG in the nonconvex settings, we show that the proposed methods converge, targeting stationary points of an explicit global functional. Experimental evaluations conducted on various Poisson inverse problems validate the convergence results and showcase effective restoration performance.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to apply the Plug - and - Play (PnP) method in the Poisson inverse problem for image restoration. Specifically, the traditional PnP method relies on data fidelity terms with Lipschitz gradients or closed - form proximal operators, which are not suitable for the Poisson inverse problem. The paper proposes a PnP method based on Bregman divergence to adapt to the Poisson noise model, thus effectively solving the image restoration challenges in the Poisson inverse problem. ### Background of the Paper In many practical application scenarios, such as positron emission tomography (PET) or astronomical CCD camera imaging, image data are obtained by particle counting, and the Poisson noise model is usually used to describe this imaging process. The data fidelity term under the Poisson noise model is neither smooth nor has a closed - form proximal operator, which limits the application of the traditional PnP method. ### Main Contributions 1. **Introduction of Bregman Divergence**: - The authors propose using Bregman divergence instead of the Euclidean distance to better capture the smoothness characteristics of the problem. - Through Bregman divergence, the Poisson noise model can be more appropriately handled. 2. **Proposal of Bregman Score Denoiser**: - A new denoiser, the Bregman Score Denoiser, is introduced, which is specifically parameterized and trained for the new Bregman geometry. - It is proved that the Bregman Score Denoiser corresponds to the proximal operator of a non - convex potential function. 3. **Algorithm Design**: - Two PnP algorithms based on the Bregman Score Denoiser are proposed to solve the Poisson inverse problem. - Based on the Bregman Proximal Gradient (BPG) method, its convergence results in non - convex settings are extended, and the convergence of the proposed method is proved. ### Experimental Verification - Through experimental evaluations on various Poisson inverse problems, the convergence and effective restoration performance of the proposed method are verified. ### Formula Summary - **Poisson Data Fidelity Term**: \[ f(x)=\sum_{i = 1}^m\left(y_i\log\frac{y_i}{\alpha(Ax)_i}+\alpha(Ax)_i - y_i\right) \] - **Bregman Divergence**: \[ D_h(x,y)=\begin{cases} h(x)-h(y)-\langle\nabla h(y),x - y\rangle&\text{if }x\in\text{dom}(h)\\ +\infty&\text{otherwise} \end{cases} \] - **Bregman Score Denoiser**: \[ B_\gamma(y)=y-(\nabla^2h(y))^{-1}\cdot\nabla g_\gamma(y) \] ### Conclusion The paper successfully introduces Bregman divergence into the PnP method and solves the image restoration challenges in the Poisson inverse problem. By introducing the Bregman Score Denoiser, not only the applicability and effectiveness of the algorithm are improved, but also the convergence of the algorithm is proved. The experimental results further verify the effectiveness of the proposed method.

Convergent Bregman Plug-and-Play Image Restoration for Poisson Inverse Problems

Convergent Plug-and-Play with Proximal Denoiser and Unconstrained Regularization Parameter

Single-Shot Plug-and-Play Methods for Inverse Problems

Generative Plug and Play: Posterior Sampling for Inverse Problems

Plug-and-play Inertial Forward–backward Algorithm for Poisson Image Deconvolution

Convergent regularization in inverse problems and linear plug-and-play denoisers

Provably Convergent Plug-and-Play Quasi-Newton Methods

Constrained Plug-and-Play Priors for Image Restoration

Block Coordinate Plug-and-Play Methods for Blind Inverse Problems

Deep Block Proximal Linearised Minimisation Algorithm for Non-convex Inverse Problems

Plug-and-Play image restoration with Stochastic deNOising REgularization

Tuning-free Plug-and-Play Proximal Algorithm for Inverse Imaging Problems

A Unified Plug-and-Play Algorithm with Projected Landweber Operator for Split Convex Feasibility Problems

Convergence Analysis of a Proximal Stochastic Denoising Regularization Algorithm

Plug-and-Play Image Reconstruction Is a Convergent Regularization Method

TFPnP: Tuning-free Plug-and-Play Proximal Algorithms with Applications to Inverse Imaging Problems

TFPnP: Tuning-free Plug-and-Play Proximal Algorithm with Applications to Inverse Imaging Problems

Fast roughness minimizing image restoration under mixed Poisson-Gaussian noise

On the Strong Convexity of PnP Regularization Using Linear Denoisers