Yves,
I'd like to know the answer too and I know it can be complex.
First, take the 24 measured colors for which we have XYZs for.
On the other hand, we have the camera's response to those colors, for which we
have RGBs for.
If we have to calculate the relationship between "one" color, we would have one
set of RGB color on the right side of the equation and corresponding XYZ color
on the left side of the equation.
Then our task becomes to find a set of 'coefficients' that, when multiplied
with each RGB value, will result in the measured XYZ value. Simple? Now, repeat
the experiment for the 23 other colors.
Can you see that, it then becomes a matter of "averaging" the coefficients?
I'm sure it's more complex than that but that's one approach.
In the end, if we're building a simple Matrix ICC profile, we're only allowed 9
coefficients to transform camera RGB to measured XYZ.
There is nothing that guarantees that the "fit" will be perfect.
I confess I don't know enough about popular mathematical models behind camera
profiling.
With argyll, at least, there is a choice of "LUT-based" ICC profile instead of
the simpler Matrix one, a few others, with TRCs.
I can't say whether a LUT-based digital camera profile is "superior" or a
"better fit" than a Matrix model?
I never compared the two on the base of DeltaEs or visual smoothness -- shame
on me ☹
But I suspect the majority of camera / scanner profilers out there stick with a
linear matrix model.
/ Roger
-----Original Message-----
From: argyllcms-bounce@xxxxxxxxxxxxx <argyllcms-bounce@xxxxxxxxxxxxx> On Behalf
Of Yves Gauvreau
Sent: Friday, February 26, 2021 9:25 AM
To: argyllcms@xxxxxxxxxxxxx
Subject: [argyllcms] Scanin question
Hi,
I'm trying to create a simple matrix profile for my camera. I'm using a
ColorChecker Passport 2 for this preliminary profile. I'd like to establish a
good workflow before getting over my head.
To make a long story short, I see the values for the patches in the ti3 file
have nothing to do with the 16 bits values I can see for the patches in
Photoshop lets say.
How those original 16 bit values get converted to those I see in the ti3 file,
I'd like to know if possible?
Thanks,
Yves
