"Right" color. How?

. There is a point of view that the type of electrical charge transfer mechanism in sensors, CCD or CMOS, affects the color characteristics of the resulting image. Adhering to this point of view, I nevertheless tried to get the same color by using cameras with these two types of sensors. Since the overall result was unsatisfactory, I continued to stick with that viewpoint. However, the results of a blind test, in which I guessed correctly only about 50% of the time, convinced me if not of the fallacy, then of the doubtfulness of this statement.
Also, my doubts were reinforced by the fact that the cameras with "correct color" included cameras with Foveon sensors, which are not only CMOS sensors too, but having very poor color separation and requiring very aggressive color reconstruction essentially mathematical coloring of the picture. At the same time, all of this led to the idea of the presented solution.
Another justification for the unsatisfactory results of the blind tests was also taken into account: "images from CCD cameras are not properly processed", i.e. "corrupted" to the level of CMOS images.
In each pair, one shot was taken with the "correct" camera (S5Pro, Foveon and the like...) and processed in a converter with default settings except for white balance.
That is, it was not subjected to any influence from the photographer and it was not "corrupted" in any way by the processing. It's pure machine work.
On the test page you can try to guess which camera took which picture.
All conditions: ISO 100 to 200, same CZ Distagon 25/2.8 ZF lens (except Foveon), same aperture. Please refer to the EXIF files for details.

Methodology

We will try to get the color of the target camera in an image taken with a test camera. To do this we first build a model of the test camera in RGB target space:

photograph the target IT8 with both cameras in raw format
process both photos in C1Pro converter, which lets you output the result in TIFF format of 48 bits per pixel with embed camera profile, so without CMS conversion and in linear responce mode, so without applying film curve.
measure target components
using Principal component analysis - PCA method, we build the trial chamber model and get the RGB value conversion coefficients from the trial chamber space to the target space.

For the test shots we do the following:

process the image in the C1Pro converter with embed camera profile, that is, without CMS conversion and in linear responce mode, that is, without applying a film-like curve.
apply the obtained transformation
assign a color profile to the target camera
apply a film-like curve to it, like the one the converter uses.

As mentioned above, the converter processes the target camera image in the default setting. The shots taken by Sigma DP2s were processed in SPP, the rest in C1Pro.
Other than the processing described above, no manual processing of the images from the test camera was done.
This approach circumvents two problems:

no matter how inaccurate the target camera profile is in the colorimetric sense, if the two resulting images look the same, then the RGB components before the CMS conversion are also the same, and there is no difference in color transfer ability between the cameras as such
no matter how inaccurate the target itself is about the color data of its fields supplied with it, and no matter how much it degrades over time, if the images are taken by both cameras at the same time, it does not play a significant role.

The quality of the resulting models is characterized by the determination coefficient R² and is given below.

Results

Please note: The image in the right column depends on the cursor position. Watch your cursor position for proper viewing!
Three shots with Fuji S5Pro(CCD), Sigma DP2s(Foveon) and Sony A7r(CMOS) cameras, adapting A7r to the first two

Fuji S5Pro	Sony A7r R²=0.999 Hover mouse to view original

Sigma DP2s Foveon	Sony A7r R²=0.985 Hover mouse to view original

Two shots taken with Fuji Velvia 100 film and a Nikon D70 digital camera

Fuji Velvia 100, Nikon F80, Nikon Super Coolscan 4000 ED, part of the frame	Nikon D70 R²=0.977 Hover mouse to view original	Nikon D70 DxO FilmPack

Now let's try to solve the previous problem

Nikon D70	Cannon 20D R²=0.993 Hover mouse to view original

Although we don't have the camera models described by there, assuming that the Nikon D50 can have the same filters as the Nikon D70 , and the Canon 20D as the Canon 30D , let's try to apply the previous model to these pictures

Nikon D50	Cannon 30D Move your mouse to view the original

Body tone test.
We also do not have a Nikon D800, but we will try to use a Sony A7r camera model, based on the possible similarity of the filters of these cameras.
Photos courtesy of Zebra, a member of the IXBT.com Conference. Digital Photography

Fuji S5Pro	Nikon D800 Hover mouse to view original

Fuji S5Pro	Nikon D800 Hover mouse to view original

Examples

Fuji S5Pro	Sony A7r R²=0.999 Hover mouse to view original

Fuji S5Pro	Sony A7r R²=0.999 Hover mouse to view original

Test at low light source color temperature - incandescent lighting
Closest Planckian temperature = 2818K (Delta E 0.178814)
Closest Daylight temperature = 2763K (Delta E 0.351169)

Fuji S5Pro	Sony A7r R²=0.999 Hover mouse to view original

Test at low color temperature of the light source - fluorescent lighting
Closest Planckian temperature = 2474K (Delta E 9.449833)
Closest Daylight temperature = 2244K (Delta E 14.519668)

Fuji S5Pro	Sony A7r R²=0.999 Hover mouse to view original

Sigma DP2s Foveon	Sony A7r CZ Planar 50/1.4 R²=0.985 Hover your mouse to view the original

I never liked the yellowish tint of green on the Nikon D70, that's also a solvable problem

Fuji S5Pro	Nikon D70 R²=0.999 Hover the mouse to see the original

Two shots taken with Fuji RealaSuperia 100 film and a Nikon D70 digital camera

Fuji RealaSuperia 100, Nikon F80, Nikon Super Coolscan 4000 ED, part of the frame	Nikon D70 R²=0.982 Hover mouse to view original

Two shots taken with Fuji RealaSuperia 100 film and a Nikon D70 digital camera

Fuji RealaSuperia 100, Nikon F80, Nikon Super Coolscan 4000 ED, part of the frame	Nikon D70 R²=0.982 Hover mouse to view original

Conversions between converters

It's no secret that converters produce different conversions for the same RAW file. As you can see from this research, this is due to the different color profiles used by these converters. This section illustrates how one converter can convert from one color to another.
The Fuji S5Pro image already shown above

Fuji HU3 converter	Converter C1Pro converted to Fuji HU3, R²=0.996

The conversion could not fully compensate for the peculiarities of the HU3 color profile, which gives the sky a greenish hue.
Test of flesh tones.
We also don't have a Nikon D800 camera, but we will try to use the Sony A7r camera model, based on the possible similarity of the filters of these cameras.
Conversion from A7r to S5Pro + C1Pro followed by conversion to S5Pro + HU3 Images courtesy of Zebra, a member of the IXBT.com Conference. Digital Photography

Fuji S5Pro	Nikon D800 Hover mouse to view original

Conversion, especially double conversion, and from a camera supposedly similar to the Nikon D800 doesn't give good color similarity.

Problems

In the wavelength range shorter than 460 nm, the Fuji S5Pro has only the blue channel of any sensitivity.

Source Messung der spektralen Empfindlichkeit von digitalen Kameras mit Interferenzfiltern, p. 75.

This means that all colors in this range will be rendered in gradations of blue. The model is unable to reproduce this flaw in the S5Pro.

The color of the 405nm laser spot.	Sony A7r converted to Fuji S5pro

Another problem is the sensitivity of the matrices to infrared radiation.
As you can see the S5Pro is more sensitive to IR than the A7r. This could explain the red color distortion in incandescent photos that is not present in fluorescent photos, despite the close color temperatures of the light sources. In fact, the S5Pro was experiencing severe stray light in the IR range.
The pictures were taken through an IR filter with a bandwidth of 720 nm and longer.

Fuji S5Pro, ISO 160, F 2.8, 2.2 seconds	Sony A7r, ISO 100, F 2.8, 30 seconds

Other tasks

. Using this technique to create a test RAW converter for Sigma DP2s (Foveon). Click on image to view