The paper you cited chose to analyze the Van Gogh paintings by picking a spot on Starry Night, measuring the color, then moving 6 pixels, and measuring the color again, then moving 6 pixels… etc. then they looked at the probability distribution of the color data they collected, then did it again only at 20 pixel distance between measurements, then again at 40 pixel distance between measurements and again at 60 and then 80 pixel distances. They found almost the same probability distributions at every distance (the distributions at 6 pixels were not exact). Do you know what the distributions were? They were normal, just normal. The probability distributions on the Kolmogorov fluid turbulence probability distribution functions were also normal (at least when viscosity is orders of magnitude different from energy in the “drops” of fluid). The reason it didn’t work on the self portrait is that it’s not a sky with a bunch of similar but slightly different colors.