It is possible to derive affine transformations that take some image and scale, rotate, skew and translate it.

These copies should be contractive and provide good coverage, so that if we overlay them, the image we obtain looks like the original image.

If we then take some approximation to an image, we can apply the transformations to this approximation, overlay all the transformed images and obtain a better approximation.

Both the random and growth algorithms use a single painted pixel as the initial approximation. If we then apply any number of transformations to this pixel's co-ordinates, then overlaying this transformed painted pixel gives us a better approximation to the original image. This gradually paints pixels that should be painted, one extra per applied transform until we get back to a previously painted pixel.

The Death algorithm begins with a painted square or rectangle that we know contains the original image. We then apply the transformation to each pixel's co-ordinates and unpaint those co-ordinates that are not mapped to. This gradually unpaints pixels that should not be painted.