Hue Twists
Whenever you mention color in photography and the differences between cameras, people always say “Just use a Color Checker.”
A Color Checker is a series of swatches meant to represent commonly photographed objects.
There are patches for RGB, CMY, the sky, plants, skin tones and a few other carefully chosen things.
If you profile your camera using something like a Color Checker, you get something like this. A slice of a color wheel with a few dots on it. This is the Color Checker profile for my personal Canon 5D mk2.
It leaves large areas relatively untouched. It ignores a large portion of green, for example.
The color checker is like a circular board with pegs sticking out of it. The fabric of the color space — the data in the RAW file — is then stretched so that the corresponding spots in the RAW file are mapped to the corresponding spots on the board.
This is what the dots and arrows correspond to. One is the correct peg location, and the other is the location on the fabric mapped to it, but I forget which is which.
Which means all in between the dots, the corrections are stretched in hue and saturation space in one direction or another, by some algorithm that’s unknown to me.
Take the 3 dots in a row in the approximately 2:00 position. This is the skin tone line. Colors in this this line correspond to a certain type of melanin produced in the skin. Notice how the middle dot of the three is being pulled inward (or is it outward?) — same hue, different saturation.
Side note: There are two types of melanin in the skin (Eumelanin and Pheomelanin) and more types in the eyes and hair. The dominant (caucasian / black) type of melanin is along this line — the more melanin, the more “saturated” the skin. The other type of melanin (mostly seen in Asiatic peoples) is a few degrees counter clockwise and is likely represented by the other dot closer to the 1:00 position. Camera companies need to carefully balance hue and saturation along this line to produce good skin tones. While light skin tones may benefit from a bit of saturation, too much saturation will produce a cartoonish over-saturated look in more melanin rich skins. This may explain the deliberate hue shifts in Canon’s colors versus the Color Checker ‘neutral’.
Of course the color wheel is actually just a slice of a 3 dimensional color pie, it leaves out Value.
If Hue is which color, and Saturation is “how much white is added” (or how much hue is removed), Value is “how much black is added”, such that the bottom layer of the “cake” is pure black, and the top is pure hue. The whole central column is grey, with the top center dot being pure white.
Value then maps to something like Luminance, which is something like Exposure. (Sorry for the vague terminology, I’m too lazy too look the definitions up and I’m sure someone will correct me if I say they are actually the same.)
Thanks to the work of Sandy at the excellent ChromasSoft blog we can see how different camera JPG engines change Hue and Saturation as Value changes.
Below are representations of the Standard and Portrait camera profiles from Adobe for the Canon 5D mk2 in the HSV space. I took Sandy’s values and plotted them in 3 dimensional space using Apple Grapher.
The color disc in the below image represents the bottom of the HSV space and should be black nor are the dots rendered in the correct HSV space, it’s a limitation of the software that “Value” isn’t taken into account when rendering color.
I suspect she basically took photos of a color checker, and mapped the resulting camera JPG to the appropriate dot in HSV space.
As you can see, the dots “twist” as Value increases or decreases.
So for some reason, Canon chose to make their sensors not “accurate” according to a Color Checker*, and then chose for their JPG engine to not treat all Values as the same in Hue and Saturation space. As Value increases, Hue and Saturation “twist”.
* a vast oversimplification
Some colors “twist” in more extreme ways than others. This is all part of Canon’s intent for what “Standard Color” and “Portrait Color” are. In the videography world one might call the formula for mapping RAW values to HSV values a LUT or Look Up Table.
When someone talks about color, we tell them to take a Color Checker — which has a few set dots within HSV space and stretches the RGB values from RAW data (that is, the red or green or blue exposed sensels) to the appropriate location in HSV space.
It’s a bit of a mystery whether or not the resulting Color Checker calibrated RAW conversion contain any hue twists (though one could easily reproduce this test with a Color Checker and Adobe).
So — whenever someone says to me “Just use a Color Checker and your camera will be perfectly calibrated” I say…
What about the specifics of the crossovers between the Red, Green and Blue dyes in the CFA (color filter array)? That is — what if some subtle gradations of colors aren’t even picked up by the sensors?
What about Metameric Failure where, as a result of the dyes specified in the color filter array, two or more sets of real world frequencies of light map to the same RGB values on the sensor? The sensor cannot distinguish between two different real world “colors” as the eye might perceive them, so which gets rendered?
What about all the spaces in between the dots?
And what about hue twists? An obvious and common one I see lately is “make darker values more blue, make lighter values more orange” creating within the same photo an “orange and teal” look, regardless of the subject. But why is it used other than for obvious effect? What about when it’s used for subtle effect? What about when it’s used to improve color accuracy, because the eye itself “twists” hue as value changes?
Canon certainly chose those hue twists for a reason. Why is this never talked about?
Color is a nuanced and difficult concept to fully wrap your head around. It’s certainly more complex than “megapixels” and “high ISO noise”, and perhaps that’s why our conversation around color isn’t as nuanced as it could be.
