This I completely agree with, but for somewhat different reasons.
Images source: Rogers, D. (2006). The Chemistry of Photography, 109-110.
Not even close, but is this really relevant? The scene (which is what the the captured light has been emitted from) is not part of the flow anymore. The negative is the scene, and the backlight now is the spectrum that needs to be accounted for.
Then this conversion workflow is not repeatable or consistent, which doesn’t make it very useful. Yet, we see in the analogue process that printing (and scanning!) of color negatives can be very repeatable and consistent.
Yes, but there is more to it.
A (negative) image on film is a component of the analogue process. What is the analogue process after the negative has been developed? It is printing on paper using an enlarger. The enlarger projects the image using a tungsten-halogen light source. The characteristics of that light is the temperature of around 3800K and the spectral composition not too far away from a 5500K black body, or the sunlight. The light has a yellowish whitepoint but other than that is neutral. That can be considered a reference light source in which the negative looks correct… to what? To the paper.
Like was mentioned on these forums elsewhere, the paper color response is what represents the observer. That is, the colors on the negative cannot be interpreted correctly by a “Standard Observer”-based system, such as a human (obviously), or a digital camera. The colors need conversion, which is implemented by the paper chemistry and (!) takes into account any specifics of the film’s color reproduction. One of those is, naturally, the color of the unexposed film areas, others are whatever color channel irregularities that may be there, which are not likely to be linear or evenly distributed (which comprises a ‘look’). Here is a paper spectral sensitivity chart, note how the channels response is proportional to the channels density on film:
Image source: Fuji Crystal Archive Paper Data Sheet, p.3.
How can we emulate the same in digital? One way is to have a paper-like ICC profile which, by simply being applied to a negative would transform it to a nearly balanced positive. If necessary, an additional CMY color balance global correction would bring it to a completely balanced state. This is analogous to what the enlarger color knobs do.
Is this enough? For one film/paper combination - yes, but what about the others?
Then there is the question of the ‘look’, those same irregularities (compared to a linear response) in the channels (or, more precisely, the emulsion’s spectral response), which we generally would like to preserve. A transformation algorithm, based on some key values (such as WB, WP, BP, and per-channel gamma) would not necessarily give adequate results, because the channels may not have those assumed relationships.
One such discrepancy which I still don’t have a solid understanding of is the behaviour of the orange mask. It is described as a dynamic system, producing a decreasing proportion of the orange mask color with increasing exposure. The cyan layer is the topmost and is placed above the mask, so the red channel response is unaffected by this behaviour. The green and blue channels each do gain extra input from the couplers, increasing with the exposure. As far as I understand, in practice this means a dark green patch on the negative would have more yellow in it (coming from the unreacted coupler) than a light green one of the same (exposure spectrum) hue, and similar for the blue channel. This should place the green and blue curves at an angle to the red one, but it doesn’t on the Figure 2 chart, so it is not quite completely clear. Here is an excerpt from Rogers, D. (2006). The Chemistry of Photography, 109-115.
The Chemistry of Color excerpt.pdf (3.8 MB)
Nevertheless, the same Figure 2 chart demonstrates (and what is explicitly mentioned in the text) that on the negative the three channel response curves are approximately parallel. Meaning a simple per-channel exposure compensation would bring them pretty close to neutrality.
This I believe is what is mentioned as Integrating to Grey in Hunt, R.W.G. (2004). The Reproduction of Colour, 44-45:
Getting back to the point, reproducing the negative on scan as faithfully as possible indeed seems to be important. White balance and even the ICC profile should be applied to the backlight.
From there the promising direction seems to be very different from the current FilmNeg approach. There were indeed some very pleasing conversions posted in this thread. Yet, in complex cases where there is no neutral points FilmNeg struggles, while a decades-old manual PhotoShop channel equalization approach works very well indeed. Maybe it just needs to be automated.