1: \begin{abstract}

2: Variational models with coupling terms are becoming increasingly

3: popular in image analysis. They involve auxiliary variables, such

4: that their energy minimisation splits into multiple fractional

5: steps that can be solved easier and more efficiently. In our paper

6: we show that coupling models offer a number of interesting properties

7: that go far beyond their obvious numerical benefits.\\

8: We demonstrate that discontinuity-preserving denoising can be

9: achieved even with quadratic data and smoothness terms, provided

10: that the coupling term involves the $L^1$ norm.

11: We show that such an $L^1$ coupling term provides additional

12: information as a powerful edge detector that has remained

13: unexplored so far.\\

14: While coupling models in the literature approximate higher

15: order\linebreak

16:  regularisation, we argue that already first order coupling

17: models can be useful. As a specific example, we present a first

18: order coupling model that outperforms classical TV regularisation.

19: It also establishes a theoretical connection between

20: TV regularisation and the Mumford-Shah segmentation approach.

21: Unlike other Mumford-Shah algorithms, it is a strictly convex

22: approximation, for which we can guarantee convergence of a

23: split Bregman algorithm.

24: 

25: \end{abstract}

26: