1: \begin{abstract}

2: Much research has been devoted to the problem of restoring

3: Poissonian images, namely for medical and astronomical applications.

4: However, the restoration of these images using state-of-the-art

5: regularizers (such as those based on multiscale representations or total

6: variation) is still an active research area, since the associated optimization

7: problems are quite challenging. In this paper, we propose an

8: approach to deconvolving Poissonian images, which is based on an alternating

9: direction optimization method. The standard regularization (or maximum a posteriori)

10: restoration criterion, which combines the Poisson log-likelihood with a

11: (non-smooth) convex regularizer (log-prior), leads to hard

12: optimization problems: the log-likelihood is non-quadratic and

13: non-separable, the regularizer is non-smooth, and there is a

14: non-negativity constraint. Using standard convex analysis tools,

15: we present sufficient conditions for existence and uniqueness of

16: solutions of these optimization problems, for several types of

17: regularizers: total-variation, frame-based analysis, and frame-

18: based synthesis. We attack these problems with an instance

19: of the alternating direction method of multipliers (ADMM), which

20: belongs to the family of augmented Lagrangian algorithms. We study

21: sufficient conditions for convergence and show that these are

22: satisfied, either under total-variation or frame-based (analysis and

23: synthesis) regularization.

24: The resulting algorithms are shown to outperform alternative

25: state-of-the-art methods, both in terms of speed and restoration

26: accuracy.

27: \end{abstract}

28: