This release achieves better results by matching the target tiles with input tiles in a random order. This avoids the output appearing blocky and also ensures that the quality of the match is distributed randomly over the whole image, instead of decreasing from top left to bottom right.
Furthermore, this release removes an attempt at a parallel implementation of nearest neighbours matching, since it was incorrect. This now finally means that the number of duplicates is properly enforced and also that the algorithm produces better results as it does not make mistakes due to race conditions.