1: \begin{abstract}           

2: In this paper we are concerned with fully automatic and

3: locally adaptive estimation of functions in a ``signal $+$ noise''-model where the regression

4: function may additionally be blurred by a linear operator, e.g. by a

5: convolution. To this end, we introduce a general class of \emph{statistical

6: multiresolution estimators} and develop an algorithmic framework for computing those. By

7: this we mean estimators that are defined as solutions of convex optimization problems with

8: $\ell_\infty$-type constraints. We employ a combination of the alternating

9: direction method of multipliers with Dykstra's algorithm for

10: computing orthogonal projections onto intersections of convex sets and prove

11: numerical convergence. The capability of the proposed method is illustrated by

12: various examples from imaging and signal detection.  

13: \end{abstract}