Align R and B to G channel in trichrome merging?

[matthoosier]

I'm just wondering whether there's any particular reason that the RGB triplet image merging (added in #309) holds the R channel exposure steady spatially and shifts the G and B to align with it.

Most cameras have significantly better spatial resolution in green on their color filter arrays, and human vision is usually described as more sensitive to luminance and sharpness in green than in red/blue.

Maybe it makes sense to align R and B to G instead? Might lead to better sharpness in the result.

[seanharding]

I don't honestly know how much difference this makes in practice, but here's a quick change to try if anyone wants to test it: https://github.com/seanharding/NegPy/tree/feat/trichrome-align-to-green

[marcinz606]

what you suggest makes sense and it's worth testing. There is no particular reason why it was R-aligned initially other than R->G->B order being somewhat intuitive :D

[matthoosier]

I loaded a couple of pictures up and to be honest can't tell a lot of difference. It's an earth tones landscape type of photo, so lots of green from trees. Note, the film was very basic, and was shot about 30 years ago in a point and shoot camera by a casual user.

But nothing obvious broke, and it does seem like the right move from an information-theoretic point of view.

I don't have any extremely high quality scans to try this out. Maybe the advantages are clearer on something that's higher res and on better stock. Does anybody have high quality R/G/B scans they'd be willing to lend to do some comparisons of exports using both methods?