[argyllcms] Re: has anyone tried ColorVision PrintFIX ?
- From: Gerhard Fuernkranz <nospam456@xxxxxx>
- To: argyllcms@xxxxxxxxxxxxx
- Date: Wed, 12 Oct 2005 23:39:31 +0200
Graeme Gill schrieb:
Very possibly, but it's up to the manufacture as to what sort of light
source, and what sort of phosphor to use, so between the light and
the filters, it is possible to get it to work reasonably. When
you're using LEDs, it's pretty hard to avoid it being narrow band.
Btw, I found some old, archived experimetal results. A couple of years
ago, I had attempted to reconstruct the spectral response of my old
scanner (may god rest his soul) from spectral measurements of an IT
8.7/2 target. For the recovery I did make the assumption that the
unknown CCFL light source can be modeled as weighted sum of the twelve
fluorescent F<n> illuminants (with the weights being estimated as well),
and I did impose non-negativity and smoothness constraints on the shapes
of the sensor responses in order to obtain reasonable results (as you
know, the recovery problem is very ill-conditioned and extremely
sensitive to noise).
Graeme, sure, no doubt, that's pretty hard.
Estimated RGB sensor sensitivities (i.e. sensor+filters):
(36 bands @ 10nm)
Estimated overall response (flurescent light source+sensors+filters):
(36 bands @ 10nm)
Since I don't know the actual response, there was no way for me to
verify the accuracy of the reconstruction. One problem is indeed, that
different parameters used for the estimation (for instance different
error/smoothness trade-offs) eventually result in noticeably different
shapes of the estimated sensitivities. But I think, the results look at
least reasonable, though they are surely not perfectly accurate.
As you can see, particularly the red and green responses seem to be
relatively narrow-band too. Actually I do not have the impression, that
the manufacturer did make any attempts to optimized lamp & filters in
order to approximate a colorimetric response. If they are optimized at
all, then likely rather for other objectives (e.g. for max. light ouput
of the lamp to reduce SNR, etc.).
When I later got my Spectrocam, I also measured the spectrum of the
scanner's CCFL lamp. Here is the measurement result:
(Spectrocam has 5nm interval, 20nm bandwidth)
Other related posts: