If you're using Python anyway, you can use scipy's triangulation module to avoid having to compile triangle:<p><a href="https://docs.scipy.org/doc/scipy-0.14.0/reference/generated/scipy.spatial.Delaunay.html" rel="nofollow">https://docs.scipy.org/doc/scipy-0.14.0/reference/generated/...</a><p>The triangulation can then be plotted fairly straightforwardly with matplotlib's triangulation routines: matplotlib.pyplot.triplot and matplotlib.pyplot.tripcolor (<a href="https://matplotlib.org/examples/pylab_examples/tripcolor_demo.html" rel="nofollow">https://matplotlib.org/examples/pylab_examples/tripcolor_dem...</a>).
I did some work [1] on constrained Delaunay triangulations similar to this where visually important, and constrained, edges were automatically determined by Laplace-like filters.<p>[1] <a href="https://femtondev.wordpress.com/" rel="nofollow">https://femtondev.wordpress.com/</a>
It seems to me that the Voronoi diagram [0], the counterpart of Delaunay triangulation, is probably more useful as a general approach to this idea.<p>The cellular nature is great for lots of applications.<p>[0] <a href="https://en.m.wikipedia.org/wiki/Voronoi_diagram" rel="nofollow">https://en.m.wikipedia.org/wiki/Voronoi_diagram</a>
As suggested in the discussion points, thin lines/stripes get muddled up with wider adjacent surfaces. Is this a problem with the goal of the exercise ? Maybe it needs to be more clearly defined. I suppose the idea is that the method and the style fit together nicely, and as a result the paintings are representable quite efficiently with low resolution triangulations. It already mostly fits the bill, except for some exceptions to the main shapes represented. Does the author suggest further work should be devoted to improving the efficiency of the depiction? I guess the thin lines are quite important in keeping the paintings structured but I'm not sure what's to gain from that exercise if straying too far from the original triangulation — I think you'd lose the amusing coincidence of these concepts applying nicely together.
The other day I was looking at this coding challenge to automatically select points and create a Voronoi of an image: <a href="https://codegolf.stackexchange.com/questions/50299/draw-an-image-as-a-voronoi-map" rel="nofollow">https://codegolf.stackexchange.com/questions/50299/draw-an-i...</a><p>Of course half the fun here is the coincidence of the last names :)
Interesting experiment! Per the talking points, I think a good color strategy would be to sample the lightest and darkest points from each region, and then use them as endpoints in a gradient fill.