Why do gray and white cards give different white balance readings in Camera Raw?

Question

I photographed a calibrated gray card and white card and used the white balance eyedropper in Adobe Camera Raw. Picking the white card gives about 3950K, while picking the gray card gives about 4300K, and this happens under phone LED light, another warmer light, and daylight.

The cards are intended to be spectrally flat, and when I white balance from the gray card, the RGB channels on the gray card align closely. Why would the white and gray cards produce different Kelvin readings? Is there a technical explanation related to the camera, raw processing, or lighting, and is there a best practice if I want the most accurate white balance?

Steven Kersting · Answer

I am not interested in an artistic feeling but in physically correct measurements/white balancing

Unfortunately, this is not really possible. The problem is that the camera does not actually see/record colors; it only records light intensity after some of the light is blocked by a color filter. And then those intensity values are converted into color using white balance multipliers (green is always 1x).

For example:

Tungsten: 1.392498 1.000000 2.375114
Daylight: 2.132483 1.000000 1.480864
Shade: 2.531894 1.000000 1.223749

The specific multipliers varies by camera, and the results vary based on how accurate the original recording is (filter accuracy; it's not), and the multiplier accuracy (also not entirely accurate). I.e. the red, green, and blues used to reproduce/display the scene are different from the ones that were in the scene, and different from the ones used to record the scene.

So even if you color calibrate to make a given color reproduction as accurate as possible, that correction will make some other color less accurate. And this is essentially what you are attempting to do with a white balance correction.

Some multiplication must be applied to a raw file, and those already modified values are what are being used to set the white balance from, even for a raw file. You basically risk compounding errors for some colors. To change the white balance of the original raw rendering you have to change the demosaicing algorithm/math used... by changing the color calibration/process that is being applied in camera raw.

You can make the white balance correction "better" by having the card evenly illuminated and reflecting directly back to the camera (set perpendicular), by having the card occupy more of the image (zoom in), and by sampling a larger area to set the white balance from (drag the dropper over a larger area instead of just clicking on a point). What this does is attempt to average out all of the variables. The exposure also must not push any color to clipping after multiplication.

But in the end, we don't even reproduce colors 100% accurately; what you see is different from what I see... So it's not such a big deal most of the time.

Edit:

I believe the main reason you will get different WB results from different levels of exposure (white/grey/black) is due to noise. In order to calculate the RGB values for each pixel the values from surrounding pixels are used; which means the noise (photosite variability) is part of the RGB interpolation. More noise results in less accuracy and a different result... i.e. a small color shift.

A white area has a better SNR than a grey area, and much better than a black area. And if the image was recorded using a higher ISO the differences could be even greater; because SNR degrades more rapidly with underexposure for darker regions (i.e. ISO noise is more evident in darker regions of an image).

Why do gray and white cards give different white balance readings in Camera Raw?

2 Answers

Your Answer

Related Questions