Noise Wedding

This image was generated by combining wedding photos with random noise. Per each 128 block of noise pixels, those noise pixels were optimally rearranged (according to YUV Euclidean color distance) using the Hungarian/Munkres algorithm. Munkres is very slow, O(N^3), so I had to do it in small blocks. Future work will be scheduling the blocks differently, maybe rearranging pixels in a random order instead of linear 128 pixel blocks. This picture took 40 hrs to create in OCaml. I attempted a different color schedule (Hamiltonian path and perfect permutation but the results were very similar).

Other images and source code is found here: http://churchturing.org//y/valentines/

Munkres: http://en.wikipedia.org/wiki/Hungarian_algorithm