Regarding the SOTS 35mm reference to 16mm target … I noticed that all matching/prediction attempts produced colorized tree-trunk shadows for the 16mm. (Such poor distributions are allot easier when viewed on the LCD screen I’m using, simply by changing my head position – never before thought of that as a plus.) I’ve come across something similar when using HSL color transferring between same pictures.
When the L (lightness) of the 2 pictures have differences (contrast, brightness), the transferred HS (hue,saturation) “colors” are altered by the destination’s different L, and sometimes unacceptably so. I’ve found that pre-matching the lightness of the destination to the source (very easy to do on the B&W L splits) produces a proper color transfer result.
Enough of your excellent matches show such anomalies that they cause you constant push-here-pull-there touch ups. Would some sort of R-G-B pre-normalization step, to create a better distribution for your limited samplings, save you such re-adjustments? (I would suggest using a larger sample pool, but the increased storage and processing requirements for such diminishing returns might be prohibitive.)
The latest version of the tool does better in this respect, but some anomalies are unavoidable. I will take a look at the method you’re suggesting, and see if I can incorporate it in some way.