Good morning Gerhard,
The critical word is metamerism. I think, that pointed me into the right
direction.
Thank you very much.
Hermann-Josef
--- Begin Message ---
- From: "Gerhard Fuernkranz" <nospam456@xxxxxx>
- To: <argyllcms@xxxxxxxxxxxxx>
- Date: Sun, 14 Jun 2020 22:21:03 +0200
The actual problem of input devices is ambiguity due to observer metamerism.
This means, there exist different spectra which cannot be distinguished by
the sensor (-> because they map to the same RGB values), while the standard
observer can well distinguish them as different XYZ colors. And also vice
versa.
For a typical input device, any RGB to XYZ mapping function can only map a
(small) subset of spectral space from RGB to XYZ exactly, while spectra
outside this subset can only be mapped with more or less errors. But there
can exist different RGB to XYZ mappings each of them being exact for
different subsets of spectral space.
When using a linear matrix model as RGB -> XYZ mapping function, these
subsets are eventually 3-dimensional hyperplanes/subspaces of spectral
space. For each 3D-hyperplane you can extablish a different linear RGB to
XYZ mapping. As said, each of these mappings is "correct", but only for the
spectra in the particular hyperplane.
Consequently, when you establish the RGB to XYZ mapping for an input device,
you also need to define which subset of spectral space you want to get
mapped exactly.
No worries, you do this indirectly/implicitly, by using a calibration target
whose spectral reflectances/transmittances are representative for the
spectra you want to capture eventually. Fitting the measurement data from
this target will implicitly lead to a model optimized for the target's
spectral characteristics then.
Still it can happen, that the spectral subset defined by the calibration
target is too big to allow a unique mapping - in this case you still have to
renounce mapping accuracy for some of the colors on the target. If the
target contains color samples which you never intend to capture, then you
can remove them from the training set (.ti3 file), hereby possibly
increasing mapping accuracy for other colors.
Regards,
Gehard
Am 14.06.20 um 18:54 schrieb Hermann-Josef Röser:
Good evening Gerhard,
thank you very much for your detailed explanation! This gives me a lot to
think and read about. I’ll do my best do understand all this. We will see …
Hermann-Josef
--- End Message ---
