White Balance within Color Calibration. Sony.

Software darktable
1

I am seeking advice regarding a white balance issue that has me perplexed.

I struggle to perceive colour casts so have historically followed the age old system of incorporating a white balance card in a sample image, then picking the WB from that.

This was however producing unsatisfactory results, so I invested in a Calibrite (formerly X-Rite) passport and commenced using color calibration in preference to white balance. However, this change seems to have made little difference, and it appears to me that Calibrite based skin tones have a pronounced red cast.

I attach the Calibrite RAW and a RAW of my grandson, who obligingly placed himself in the same position immediately afterwards. I also include (otherwise identical) JPGs, based upon CC’s WB assessment (35.6 hue, 13.1% chroma), and my modified version of this (24.6 hue, 29.7% chroma). My own assessment is potentially far from perfect, but the Calibrite version seems way off.

I also include a darktable screen grab from when I calibrated CC with the Calibrite.

We are in northern Britain, where Caucasian skin is pretty white at this time of year. CC’s area mapping is set at the defaults of 50/0/0. My camera is a Sony A7Riii, which may be an issue.

It is perfectly possible that I am doing something fundamentally wrong, or my eyes are even wonkier than I imagine, but all thoughts and opinions will be most welcome.

Calibrite calibration
DSD01166.ARW (82.0 MB)
DSD01166.ARW.xmp (10.8 KB)

Screen grab
Screen grab1358Ă—781 954 KB

Headshot
DSD01171.ARW (81.7 MB)
DSD01171.ARW.xmp (14.7 KB)
DSD01171_01.ARW.xmp (15.2 KB)

DSD01171
DSD011715320Ă—7980 6.06 MB

White balance of the above is calibrated from a Calibrite (X-Rite) passport via color balance.

DSD01171_01
DSD01171_015320Ă—7980 6.16 MB

White balance of the above is my eyeballed modification of the CC/Calibrite assessment.

Edit: These files are licensed Creative Commons, By-Attribution, Share-Alike

2

One quick comment before trying to digest your comments…are you setting exposure to the value provided by CC. I think in your case for the example provided its showing 0.75 EV??

3

I have never understood the exposure value provided by CC, despite re-reading the manual and other sources many times. I have in the past re-run the calibration using various DT exposures but never found that it makes any noticeable difference. Whenever I bracket the actual calibration exposures in camera there will always be one that shows a marginally improved delta value, but it’s usually hair-splitting territory.

Moreover, I was historically seeing similar colour casts when basing my WB on a white card.

In this case, my calibration shot utilised an exposure of +.618EV and the headshot was processed using +1.60EV.

4

For what it’s worth, for me the second image has a slight bluish/greenish cast. See e.g the top of the white collar. Perhaps the mid tones are not quite where you want them (Causasian skin is about one stop lighter than middle gray).

1 Like
5

I am not the best judge, but agree that the image feels a tad cold. I can also perceive some blueness in the upper collar, but playing with with exposure and CB’s brilliance sliders doesn’t seem to impact this. Would you agree that my first image has an orange cast?

6

For the kid, yes, but the collar looks neutral in the first one. But the first one is also a bit too contrasty for my liking. Increase the fill intensity!

7

I believe its a value that comes from the grey ramp and it’s the value at which the colors will be accurate as assessed. Initially the exposure value in CC was an offset to be added to the current exposure but now it reflects the value that would correctly expose the chart

1 Like
8

I set out to photograph the chart. The kid was an afterthought :slightly_smiling_face:

3 Likes
9

Thanks - I wasn’t aware of that change.

At .618EV I was pretty close to the recommended .75EV, and I have just discovered that CC produces the same WB assessment as previously when I change exposure to -5EV and +5EV.

10

Well, to be honest, the first one looks excellent to me. Maybe a silly question, but - have you got a hardware calibrated computer screen? Maybe the picture is processed perfectly, but you view it on not so perfect screen :thinking:

1 Like
11

No I don’t, and I also suspect that my eyesight could be part of the problem.

That said, I view pictures from many other sources on a daily basis, and never consider them to have colour casts. I also have access to a wide variety of other devices and to my mind the darktable related red/orange cast prevails across them all. Moreover, I look after my grandchildren most days, so I know how they usually appear :grinning:

12

There is another tool or perhaps better described as a reference when you hover over numbers in the color picker you get the “name” of the color…. There are also entries for the “average” skin tones. Luminance can actually a large part of the perceived “tone” Skin tones are often pegged as salmon or crimson for Caucasian skin in this reference in DT and if you lower the exposure enough to get say below L=65 on a lab scale for the high end it will then report “average” Caucasian as the color…… you could experiment with your “eyeball” settings and try to match to the reference and see what you think…

Does the skin optimized variant work any better… Seems like your example would be a good test case and you optimized using the “none” setting…

Also as your camera is a Sony and I think a couple of others have noted improvements after doing this exercise you could try this…

https://darktable-org.github.io/dtdocs/en/module-reference/processing-modules/color-calibration/#caveats

1 Like
13

I used your Calibrite image to calibrate the portrait (I am getting the same Hue and Chroma values than you, as expected) and this is the output I am getting after working with tone equalizer and boosting the colors with Color Balance RGB…

(Please note that the as usual with pixls.us, images embedded in a post need to be clicked on to show a better version from a color and luminosity standpoint).

dt2k_DSD01171
dt2k_DSD011711280Ă—1920 243 KB

DSD01171.ARW.xmp (17.9 KB)

1 Like
14

For my two cents worth the top image of the child is better as it has warmer and more appealing skin tones. The bottom one has an unpleasant coldness to it.

Also you should be able to come up with some values for the area color mapping that approximates a nice skin tone for your subject and use that target value on the skin. Find a nice skin tone to measure for the value.

2 Likes
15

Regarding white balance one can get lost rather easily in darktable. One difficulty for example is the interaction between the white balance module (D65 matrix) and the color calibration module. I’m not experienced enough to navigate you through this topic. Just wanted to share this, because at least for my Sony A74 the darktable defaults are not well suited to give me the expected colors after white balance. :man_shrugging: (I fixed that with a custom preset for the D65 values in the white balance module)

Often it helps, to view the rendering of an other viewer/editor. So I thought I’d share this ART version for comparison reasons. I took the white balance from a grey patch of your card and did basic editing of the picture (dimmed highlights, adjusted tone curve, adjusten saturation plus a bit of local contrast). For my taste, the white balance came out rather nice.

DSD01171
DSD01171805Ă—1204 382 KB

DSD01171.ARW.arp (14.4 KB)

3 Likes
16

Thank you. I have been completely oblivious to this facility until now.

This led to me to exploring the possibility of reducing red, and has potentially put me on the right track. I have this morning experimented with reducing red through various means. To my eyes, this appears to fix the colour cast without affecting much else, but I recognise that I may be alone in perceiving this.

I attach before and after JPGs below. The upper photo utilises the WB as determined by the Calibrite, and the lower image has the same edit, albeit for an additional bypass copy of CC, which has the Red channel’s input red crudely reduced to 0.940.

DSD01171_03
DSD01171_035320Ă—7980 2.43 MB

WB as determined by calibrating with Calibrite (X-Rite) Passport

DSD01171_05
DSD01171_055320Ă—7980 2.43 MB

WB as above, but additional CC having Red channel’s input red set at 0.940

1 Like
17

Thanks for contributing this. I enjoy seeing other people’s edits and find it very informative.

18

Hah - It clearly takes a fellow Sony owner to see the red cast (and get rid of it). :partying_face:

Your edit re-assures me that I am not going crackers and that all is not lost.

1 Like
19

I have to admit that my attempt at fixing it through adjusting CC’s hue and chroma were not my most successful, but boy, you want to see how bad I can be with WB’s temperature and tint sliders.

The red cast is driving me nuts and I am amazed that others cannot perceive it. The latter includes my darling wife, who is slowly losing patience with my efforts to resolve it.

I remember AP mentioning in a video that whereas most people cannot perceive colour differences below 2.3 delta, those who do it for a living can perceive differences of 1.6 delta (or something along those lines).

20

I think the end result of the shooting a picture of your white screen to get new calibration values for D65…thought I believe calibrated is better and yours might not be… was for many Sony people a set of D65 values that did indeed correct and reduce mostly the red.

1 Like