Can you verify a D65 light source with a colorimeter or camera custom white balance?

Hardware
, , , ,
1

Continuing the discussion from Best practice for "unexpected white balance coefficients" in darktable:

In order to normalize the RGB channels of my photos prior to debayering the raw sensor data, the darktable manual recommends taking a photo of a white screen calibrated for a D65 illuminant. I have various monitors, none of them great, as well as some video light panels that let me dial in a temperature and may be better calibrated than my monitors. I’d like to figure out which source produces the best D65 photo that would let me calibrate my camera sensor. I have two ways of measuring color temperature:

  • My camera, a Sony A7IV, has a custom white balance option that lets me set white balance by capturing a photo of a light source or gray card (the results seem similar whether I point the camera at the light source or gray card). The custom white balance includes a Kelvin color temperature.
  • I have a Calibrite ColorChecker Display colorimeter, which lets me read a color temperature by running the ArgyllCMS command spotread -T.

What I’ve noticed is that when I set my video lights to 5000K, spotread -T reports about 5000K in all three temperatures. However, when I set the video lights to 6500K, or attempt to photograph a white window on my monitors (after calibrating them with displaycal and my colorimeter) the results are all over the place. The lights have a much redder temperature than 6500K (lower number K), while the displays seem to be much bluer (higher K). For example, here are some results:

  • Elgato key light, 6500K 30% power. Sony reads 5900K. Spotread says:
 Result is XYZ: 2400.654082 2515.003395 2478.452114, D50 Lab: 323.863230 -4.913590 -35.787977
                           CCT = 5852K (Duv 0.0031)
 Closest Planckian temperature = 5729K (DE2K 4.3)
 Closest Daylight temperature  = 5863K (DE2K -0.3)
  • Luxli cello 6500K 25% power, diffuser pancaked onto panel so you can still see the individual LEDs. Sony reads 5800K, spotread says:
 Result is XYZ: 2698.884049 2827.109535 3264.812617, D50 Lab: 337.377479 -5.050037 -72.305131
                           CCT = 6899K (Duv 0.0023)
 Closest Planckian temperature = 6755K (DE2K 3.3)
 Closest Daylight temperature  = 6961K (DE2K -1.3)
  • Luxli cello 6500K 25% power, diffuser expanded the way it’s supposed to be used. Sony says 5700K, spotread says:
 Result is XYZ: 2636.102874 2589.558928 2819.402903, D50 Lab: 327.188918 27.002179 -57.349215
                           CCT = 5739K (Duv -0.0080)
 Closest Planckian temperature = 6132K (DE2K -9.6)
 Closest Daylight temperature  = 6332K (DE2K -12.5)
  • Wacom cintiq pro 13HD (best of my displays), white rectangle after calibration with displaycal. Sony says 6900K, spotread says:
 Result is XYZ: 189.235728 198.537183 237.360789, D50 Lab: 129.793650 -2.410466 -33.098451
                           CCT = 7218K (Duv 0.0025)
 Closest Planckian temperature = 7041K (DE2K 3.5)
 Closest Daylight temperature  = 7271K (DE2K -1.0)

I just got my colorimeter, so don’t know that much about how to use it. There’s a dizzying array of options to spotread, so maybe I need to use one of these. (Quick experimentation suggests -e for emissive doesn’t change my readings much.) I should also say that my camera custom white balance readings seem to vary by a few hundred K depending on the lens that I use.

My questions:

  • What should I trust more, my camera or my colorimeter?
  • Which (if any) of the above sources seems like the best photo to calibrate my initial white balance in darktable, which requires a photo as close to 6502K as possible? (Note the pancaked diffuser is not really an option because you see the individual LEDs, but I included it because the temperature was so different.)
  • Short of buying a fancy monitor, is there some alternative source of 6502K light I’m overlooking or a more appropriate measurement device?

Thanks.

2

midday daylight from a clear sky.

3

Looks like all your lights are LED, which means they’re not going to present a continuous spectrum. That, and your colorimeter doesn’t measure full-spectrum, so there’ll be inexactness what it presents for color temperature evaluation.

I’d just wait for a sunny day. Then, you don’t need to measure it to assess its validity.

4

My Sony measures 5500K on a gray card under clear sky conditions. I think that unfortunately the D65 whitepoint required by darktable is significantly bluer than what you get under sunlight, so I need an artificial source.

5

I understand this, although the darktable manual does recommend photographing a screen, which is presumably also not going to present a continuous spectrum. The Luxli Cello has a CRI of >95 for what it’s worth.

I have a friend who is willing to lend me a spectrophotometer, so I could in theory characterize my LED lights better. However, I’d also consider buying a light source if I can get one for under the price of a good monitor and also use it for photos/videos.

6

@Frisco This might be of interest to you: Spectral emissions from modern light sources (more graphs added)

Have fun!
Claes in Lund, Sweden

7

Hmmm… I’d rather trust the fact that the sun and atmosphere are known constant than the three-channel-limited camera and the wonky algorithm required to assert a temperature. Unless you live in a smoggy place, then there’s no telling.

8

Isn’t temperature and tint sort of a fudge in the first place? For example, darktable and rawtherapee don’t measure the temperature the same.

9

They are a known constant, but not the constant that the early stages of the darktable processing pipeline are expecting–i.e., what happens between the white balance module and color calibration in the modern chromatic adaptation flow.

But since I can get a good 5500K picture of a gray card, is there a way to compute the D65 channel coefficients from the 5500K ones? I get R/G/B of 2.419/1.000/1593 for the gray card in the sun, 2.615/1.000/1.531 for the Elgato Keylight at 6500K with 30% power,
and 2.849/1.000/1.546 for white on my Wacom pen display. (Obviously something is wonky with these numbers, since the Wacom and Keylight should be colder than the gray card.)

Given that I don’t have a D65-calibrated display as suggested in the darktable manual, what’s the best way to compute the initial white balance coefficients (not to be confused with the color correction settings that adjust for the real white balance of a photo I’m editing).