Sorry for the delay -- had to run to campus for a meeting.
selmslie wrote:
The differences are not minor.
You misunderstood. I said the difference is minor between the X-Rite published average skin tone (hue value) for Caucasian skin and African American skin, and it is.
The operative term here is average. Of course we can sample and calculate an average and we've done that. Once we have average values they can be useful. You do understand what an average is?
selmslie wrote:
The only way you can use skin color as a guide is to base it on your own color memory having looked at the person you photographed and the color of the light with which they were lit.
No, that's wrong. Again the operative term here is average. Knowing an average skin color can be useful.
selmslie wrote:
Without that knowledge your chances of knowing whether the skin color in the photograph is right are reduced unless you have other clues in the scene.
And yes of course that information can be combined with other clues in the scene.
Here's an example of how knowing an average skin color can be useful. Let's start with this SOOC JPEG from a Nikon Z6 (thanks DPReview):
https://www.dpreview.com/sample-galleries/8048435192/nikon-z6-sample-gallery/6820584388The camera was set to auto-WB and saved the image in the sRGB color space. (The Photographer noticed the poor white balance by the way and posted a corrected version.)
The singer in the photo is an adult woman. Average Caucasian skin color in the sRGB color space has a hue value of 19. So we can start taking measurements of the woman in the photo (steering clear of make-up).
hue = 311
hue = 252
hue = 270
hue = 263
hue = 279
Those values are very far away from 19. After you do this as I have for many thousands of photos you get a sense of how far is "probably off," "definitely off," "can't possibly be right," etc. In this photo those measurement values can't possibly be accurate for the skin color not only of the singer but of the guitar player as well. Human beings simply are not those colors (at least not until some time after they're dead).
The fair assumption is an auto-WB fail by the camera. Next let's look for other clues. Is there something in the same light as the singer that we can assume is a specific color and hopefully neutral. How about the microphone. So DPReview also posted the NEF file and I downloaded that. Avoiding reflections I selected what looked like a good sample point on the mic and one click set the white balance. Let's re-measure skin color on the singer then:
hue = 20
hue = 21
hue = 19
hue = 23
hue = 22
How about that! I'm getting values that are spot on average for Caucasian skin color. And the guitar player too! And OMG! even the drummer in the back who is measuring the average skin color for an African American.
First below is two color swatches sampled from the same spot SOOC JPEG and the image white-balanced from the NEF file and then below that the two photos (resized for display here).
Do I know for a fact now that the singer, guitar player an drummer actually had average human skin color? Of course not. But knowing those average values was very useful. First they helped me to verify that the color in the SOOC JPEG was way off. By measuring the colors in the photo and comparing them to the known average I could determine there was a problem and also the way in which the colors were shifted.
Knowing that the people in the photo have average skin color once I set the white balance in the NEF file helps verify that I probably made the right assessment about the microphone. Is is likely that the people in the photo did in fact have average human skin color or skin color that was close to average?
In fact it is that's what average means and it helps that all three people in the corrected version measure as average. And having set a plausible white balance in the NEF and then getting average skin color values in all three people allows me to have a high level of confidence that I've corrected the original problem in the SOOC JPEG.
Will this be equally useful for every single photo anyone ever takes of people? Again, the operative term here is average.