Some time ago I came across this, this and this - an interesting idea to reproduce an image given a minimal set of polygons, utilising evolutionary search. The original idea used a hill climbing strategy to randomly mutate a collection of polygons, keeping a mutation only if the change yielded an improvement, defined by the sum of pixel by pixel differences between the original image and the collection of polygons in the new image.
I was curious if the method could be improved by using a genetic algorithm (using a population of candidate solutions instead of just 1) and ended up with this . Using this image as input, the objective is to evolve a new image constrained by a maximum number of polygons. Here is the result:fitness proportionate selection such that individuals with greater fitness (closer proximity to original image) have a greater chance of mating. Selected individuals then produce offspring using a genetic crossover technique and are then subject to mutation. Crossover involved copying polygons from each parent to form new offspring, while mutation involved random changes to polygon structure and colour as well as the possibility to add or remove a polygon from the newly created offspring.