This is a completely automated method build in JavaScript by akm2 (see project here: http://jsdo.it/akm2/xoYx ).
Go to the site above, and click the Play button on the right, and then drop an image into the section on that region and see what happens.
I've tried this out with generally pleasing results, though it tends to work well with pictures that are not complex – such as close up portraits.
Example: an automatically generated result
And now the source image:
The code is distributed under the MIT License so you can play with it. For example, Jeff Clark at NeoInformix has modified the code a bit to increase transparency as the triangles grow smaller: http://neoformix.com/2012/DelaunayImages2.html