1: \begin{abstract}

2:   Maximum flow (and minimum cut) algorithms have had a strong impact

3:   on computer vision. In particular, graph cuts algorithms provide a

4:   mechanism for the discrete optimization of an energy functional

5:   which has been used in a variety of applications such as image

6:   segmentation, stereo, image stitching and texture

7:   synthesis. Algorithms based on the classical formulation of max-flow

8:   defined on a graph are known to exhibit metrication artefacts in the

9:   solution.  Therefore, a recent trend has been to instead employ a

10:   spatially \emph{continuous} maximum flow (or the dual min-cut

11:   problem) in these same applications to produce solutions with no

12:   metrication errors. However, known fast continuous max-flow

13:   algorithms have no stopping criteria or have not been proved to

14:   converge.  In this work, we revisit the continuous max-flow problem

15:   and show that the analogous discrete formulation is \emph{different}

16:   from the classical max-flow problem. We then apply an appropriate

17:   combinatorial optimization technique to this combinatorial

18:   continuous max-flow (CCMF) problem to find a null-divergence

19:   solution that exhibits no metrication artefacts and may be solved

20:   exactly by a fast, efficient algorithm with provable

21:   convergence. Finally, by exhibiting the dual problem of our

22:     CCMF formulation, we clarify the fact, already proved by Nozawa in

23:     the continuous setting, that the max-flow and the total variation

24:     problems are not always equivalent.

25: 

26: \end{abstract}