By starting off with a set of random pixels, then organizing those pixels in such a way that for any given point they are the same color at the appropiate intersection points, you can fool your mind into precieving a flat image as having depth.