Originally Posted by klev
I like VSCO. What they do is actually really tough. I've actually tried to develop some color remapping. You're limited on several fronts. The available methods of color remapping typically involve enormous amounts of data, and you have a certain range of numerical precision, so safeguards have to be included to ensure something doesn't blow up under certain circumstances. Beyond that iterative methods for both edge finding and 1D mappings tend to be very computationally expensive, along with the annoying issue that they introduce certain stability issues based on the precision of the data. You can do quite a bit if you have access to properly processed hdr data, but if you want to accommodate anything down to jpegs without seeing something that looks like convergence toward a given color and with gradients along edges intact, it needs some kind of smoothing pass too. Now you're up to
find edges --> develop mapping to a suitable level of convergence for each channel --> rebuild gradients and / or introduce visually stochastic noise to mitigate the appearance of loss of precision.
Thank you for the complement, klev, a very precise explanation, IMO.