Several threads already discuss the color grading of scenes with multiple light sources (see *links with example pictures in the threads below).
The assumption: Tungsten light inside a room, mixed with limited/smaller zones with daylight from windows or a white-light-colored illuminant. Of course, this also works in night scenes of tungsten streetlights with single white LED lamps from cars.
The problem: Normal whitebalancing (i.e. using one instance of color calibration) pushes the white light into blue, partly even out of gamut.
From the paper below (Hsu et al. 2008), you’ll notice that Adobe and even the MIT are working on this problem with complex algorithms, but I’d like to try a more amateurish solution - which currently works only partially.
Note that other proposed DT solutions have focused on desaturating all blue areas with either gamut compression, color zones or rgb primaries.
I don’t know if this has already been proposed elsewhere, but my plan would be to use two instances of color calibration, with the first correcting for tungsten light, and the second one correcting for the blue zones, limiting that instance with a parametric mask.
I find this more feasible than directly including a parametric mask (to filter out white light areas) in the first and only instance - since I don’t know any way of filtering for white light (instead of any specific hue) with our current filters.
After the first instance of CC, I usually meet the following situation:
| Title: After 1st instance of Color Calibration for warm light | Originally under warm light | Originally under white light | |
|---|---|---|---|
| Gray object | gray | medium blue: correct back to gray | |
| Blue object | medium blue: do not correct back to gray | deep blue: correct back to medium blue |
The specific problem is that both a blue object in warm light AND a gray (or non-blue) object in white light end up as “medium” blue after this step.
This brings me to the challenge to fine-tuning my parametric mask: While I can limit the “blue filter” to areas with a high saturation and high luminance (to tackle at least the more extreme cases of color shifts), this corrects “fake blue” of white light areas only halfway. If, on the other hand, I set the limits of saturation and luminance lower, real blue objects (under warm light) will be affected too.
This compromise could be avoided if parametric filters could “hark back” to the moment in the pipeline both before and after the first instance of color calibration was applied, allowing me to only select only those areas not affected by the warm light (or rather, bloomed over by the white light with its broader spectrum). Of course, the current philosophy is that DT modules are regarded as isolated blocks.
Is this currently possible, would it be a desirable github request, or are there better solutions?
*Other threads with similar problems:
