He did it manually, though, which is kind of weird. Would have been cooler to automatically pick the control points by the first derivative of the image.
How does one take the first derivative of an image to choose control points?
Take a color component (say, luminance) and think of that as height. First derivative gives you a map of the "slope" between adjacent pixels, that is, how much the values differ. You get higher values where there is more complexity in the image, lower values where colors change slowly. So you lay down more control points in the "brighter" (noisier) parts of the derivative image, and space them farther apart in the "darker" (smoother) parts.
I first ran into Delaunay a few years back when I was projecting weather data. Probably the most interesting job I ever had, code-wise.
See also: fractalius
the video doesn't seem to work for me :/
Same with me. Windows 7 with Firefox 3.5 or IE 8
I use Chrome so check it off the list to :(
Makes me think of the Twitter image compression contest from a few months back. One of the entries did a Voronoi tesselation of the image.