1: \begin{abstract}

2: We propose a new fast algorithm for solving one of the standard

3: formulations of image restoration and reconstruction which consists of

4: an unconstrained optimization problem where the objective

5: includes an $\ell_2$ data-fidelity term and a non-smooth

6: regularizer. This formulation allows both wavelet-based (with

7: orthogonal or frame-based representations) regularization or

8: total-variation regularization. Our approach is based on a

9: variable splitting to obtain an equivalent constrained

10: optimization formulation, which is then addressed with an

11: augmented Lagrangian method. The proposed algorithm is an

12: instance of the so-called {\it alternating direction method

13: of multipliers}, for which convergence has been proved.

14: Experiments on a set of image restoration and reconstruction

15: benchmark problems show that the proposed algorithm is faster

16: than the current state of the art methods.

17: 

18: \end{abstract}

19: