[argyllcms] Re: Printer grey-scaling profiling inaccuracy (using scanner)
- From: Graeme Gill <graeme@xxxxxxxxxxxxx>
- To: argyllcms@xxxxxxxxxxxxx
- Date: Fri, 24 Mar 2006 23:02:15 +1100
Milton Taylor wrote:
Basically, even though the color matching seems quite good (comparing
reflective target to printout under same light), when I print a grey
scale test chart through the new printer profile, it's got a decidely
green cast in the mid section.
This is not at all unusual when using a scanner instead of
an instrument. The spectral differences between the scanner
calibration reference and the inkjet, very often cause
a color shift, and of course we are most sensitive to
errors near neutrals.
So then I had a look at what was happening in the scanner. If I scan the
IT8 target, assign the profile, all of the Lab colours are spot on
compared with the reference data, including the greyscale patches - no
chroma to speak of. So the profile itself is good.
Right, but the colorant spectrum from the photographic reference
chart, is not the same as that from the inkjet, and the scanner
spectral sensitivities are not the same as a human observer.
[Except of course that the IT8 was made on different paper with
different inks - maybe this is a factor. Just as printer profiles have
to made with the same paper and inks, does the same apply to making a
scanner profile? i.e is that profile really ink/paper specific?]
Yes. If the scanner was equivalent to a CIE observer, it would all be
fine, but scanners generally aren't. You would certainly get a closer
result if you could use a custom scanner profiling chart that was generated
on the inkjet you want to profile (catch 22 of course, since you need
an instrument to provide the reference information for the custom
scanner chart.
If I then scan grey scales made from the printer - which visually look
neutral next to the greys in the IT8 - and assign the scanner profile,
the Lab numbers are definitely not neutral. Converting to sRGB at that
point verifies that Green is down below the other two channels. Hence
this is obviously why the printer profile is boosting green.
Right. One way of improving the result would be to nudge the
scan of the inkjet chart in an editing package (say photoshop).
Since you have an injet reference that you think is subjectively
neutral, you can figure out the change to a* and b* (or
RGB) needed to make them neutral in the scan (relative to the white point).
Apply the same change to the scan of the printer profiling chart,
and then remake the profile, and it should improve.
I know there is a comment in the docs about often getting a color cast
using a flatbed to measure the targets...and I suppose this is what I'm
seeing here? But why is it happening...how can I prove it?
Your observations about the scan of the IT8 and the scan of printer
output that looks visually neutral prove it.
Is it possible to get metamerism effects with greys though? Can the
scanner look at one print, say printed with Epson K3 black ink on
Kodachrome Professional paper, and see the greys as neutral, then look
at another print made with dye-based black ink on Ilford pearl paper,
and see non-neutral greys? Even though they both look neutral to the
human eye?
Yes. It's less likely with neutrals produced using just K ink, but
it still can happen.
Secondly, what sampling algorithm does scanin use to read the patches
off the targets? Does it average or median, and typically how many
points would it take from each patch?
It takes a central region of each patch (something like 70% of
the patch area), and averages it. I think it may discard outliers
as well.
Graeme Gill.
Other related posts:
