DrDre said:
In this case the color match was done using an 8-bit color image. In practise you would use at least a 16-bit color image …
Yes, if one starts in 16-bit to finally end in 8-bit, it would fudge a solution. But the problem is, with faded film, the scanned R-G-B layers no longer have that 16-bit (or 8-bit) spread. They might have (for example, in 8-bit) only a 7 or 6 or 5-bit spread. As soon as one color-corrects it (expands it back to 8-bit), steps are introduced in that original workspace. Same for working in 16-bit.
poita said:
… add a little noise to the image. This stops the stepped-gradient effect happening when you start adjusting things. The noise can then mostly be removed afterwards.
DrDre said:
Currently, … a smoothing parameter is used to decide the smoothness of the gradient. This reduces stair stepping and similar artifacts.
My thought was to use integration – “an integral assigns numbers to functions in a way that can describe displacement, area, volume, and other concepts that arise by combining infinitesimal data” – or was it differentials (the reverse of integration; my advanced calculus is very foggy now) – to create an entirely new spread based on the old, remaining samples. It’s probably vector calculus (using both), but I never studied that.
Does this sound like a right approach (vectors) and/or what do pro-software incorporate for their solutions?