It looks like the program is rather slow. Using DotTrace , I took a look at where the bottleneck is. The method that calculates the 'fitnesse', how closely the created image resembles the original image, is where most of the time goes to. Can we improve that? So what happens in this function? We compare each pixel of both images to compare their ARGB values. The better they match, the better the result! How can be compare this? First, I tried using the naive approach: compare each pixel using a bitmaps GetPixel(x, y) . Do that however, and you'll quickly see how excruciatingly slow that is. Especially considering that even with an image of 333x333 you're already comparing a million pixels each time you compare two images! So, instead I tried using a 'FastBitmap'. It's a project I found which wraps the Bitmap and provides much faster Get and SetPixel performance. Using this instead, the performance went up by an order of magnitude! But we can still do be...
It's an old idea by now, can we recreate the Mona Lisa with fifty polygons by using random changes? The idea and original implementation by Roger Johansson can be found here . So, how does it work? You start with an source image. Then you create an empty image. We keep on doing small changes to this image. In my case this is one of these changes: Recolor our polygon Change the position of one of the polygon points Add or remove a new point to our polygon Add or remove a new polygon Switch two polygons After each action, we take a look and check if the newly changed image looks more like the original image by comparing each pixel. If it is, we continue using this one. If it's not, we discard the changes. Looks like a fun project! So I built my own version from scratch eight years ago, reusing some of the same ideas. Here's my result using the Mona Lisa: Mona Lisa with 50 polygons: You can see that it's the Mona Lisa but the details around the eyes and mou...