I recursively generated a Hilbert curve, stuck four together for a Moore curve, then did a FFT (to generate the function pair θ -> {x, y}).

Then I drew it using a filthy decompilation of some (java) image editing crap that I knocked up in uni.

I spat all of the images out to a directory (saving buffered images as PNG files) and glued them together with ffmpeg.

I'm not 100% happy with the Moore Curve, since I think that under the circumstances rotational symmetry would be more useful.

I'm also not happy with the fourier transform; I think that for something like this there should be an adequate expression for n points using log(n)/log(2) terms, but the transform that I used includes one term in the output for each in the input.

There might be a degree of something similar to "overfitting" in the result.