Re: [Lcms-user] Re: Printer-profiling via scanner

[Gerhard F. <nos...@gm...>]

Wolf Faust schrieb:

>So I see a number of advantages in building a database and trying to 
>add the spectral data of scanner, ink and everything else that is 
>easy to get/use. It's not perfect but should get us very close.
>
Wolf,

I agree, a database is important. In order that a method as described in 
http://www.ece.rochester.edu/~gsharma/papers/ntargcaljist00.pdf can be 
used *without needing a spectrometer*, a database of media 
characteristics (for various printer/settings/ink/paper combinations) 
and scanner light source spectra is actually mandatory - only scanners 
in the database can be used, and only media types in the database can be 
scanned with this or a similar a method.

If the media model parameters (e.g. spectral eigenvectors) for the 
desired medium are not available in the database, they would need to be 
measured and computed, requiring a spectrometer.

>Graeme: I agree, spectral characterization of a scanner most likely
>needs more than three dyes. I was surprised about the IEC 61966-8
>standard for spectral characterization of scanners. This standard
>uses only a target simlar to the IT 8.7 target (same layout... just
>more highly saturated colors in order to help the calculation).
>
If the goal is only the reconstruction of the sensor sensitivities, with 
methods as described e.g. in http://tinyurl.com/9mvk2 or 
http://tinyurl.com/bzyly, then I guess it is not so important that the 
calibration target covers different metamers for the same colors.

On the other hand, for the calibration of a multi spectral (more than 3 
channels) capturing system, which is expected to return the same XYZ 
values for any captured metameric pair, I also have doubts, that an IT8 
target will be sufficient, although it does not need to work for 
arbitrary spectra, but can be limited to the union of all possible 
inkjet spectra for various printers/inks/papers/..., and IT8 target 
spectra. If the union of all inkjet and IT8 spectra is decomposed into a 
reasonable number of principal components (say 9 or 12), such that all 
these spectra can be described reasonably accurate as linear 
combinations of these eigenvectors, then I also suspect, that the IT8 
spectra will utilize most, but likely not all of these eigenvectors.

Btw, does 61966-8 actually specify the recommends maths how to 
reconstruct the spectral sensitivities using this target, or is it 
rather unspecific regarding the details? Is the method similar to the 
two ones I referred above?

Regards,
Gerhard