color spaces are roughly continguous, so i suspect the algorithm doesn't matter much because the solution space doesn't have a lot of "kinks" in it where you'd

color spaces are roughly continguous, so i suspect the algorithm doesn't matter much because the solution space doesn't have a lot of "kinks" in it where you'd easily get trapped away from a global maximum. what matters more is the difference function for evaluating the colors. in this case, the author used CIE's ΔE* measurement, which attempts to take into account perceptual difference, rather than strict mathematical difference. it sounds like you could find different such functions that may subjectively perform better or worse for a given application.