1: \begin{abstract}

2: 	Image acquisition in many biomedical imaging modalities is

3: 	corrupted by Poisson noise followed by additive Gaussian

4: 	noise.   While total variation and related regularization 

5: 	methods for solving biomedical inverse problems are known

6: 	to yield high quality reconstructions in most situations,  

7: 	such methods mostly

8: 	use log-likelihood of either Gaussian or Poisson noise models, 

9: 	and rarely use mixed Poisson-Gaussian (PG) noise model.  

10: 	The work of Chouzenoux et al. 

11: 	deals with exact PG likelihood and

12: 	total variation regularization. { This method  adapts the primal-dual approach

13: 		involving gradients steps on the PG log-likelihood, with step size

14: 		limited by  the inverse of the Lipschitz constant of the gradient}.

15: 	This leads to limitations in the convergence speed.

16: 	Although     ADMM methods  do not have such step size

17: 	restrictions,  ADDM has never been applied

18: 	for this problem,  for the possible reason that  PG log-likelihood

19: 	is quite complex. 

20: 	In this paper,  we develop an ADMM based optimization for

21: 	roughness minimizing image restoration  under PG log-likelihood.

22: 	We achieve this by first developing a novel iterative method

23: 	for computing the proximal solution of PG log-likelihood,   

24: 	deriving the  termination conditions for this iterative method,

25: 	and then integrating into a provably convergent

26: 	ADMM scheme. The

27: 	effectiveness of the proposed methods is demonstrated using

28: 	restoration examples.

29: \end{abstract}

30: