Re: Why not double profile?
Re: Why not double profile?
- Subject: Re: Why not double profile?
- From: Roberto Michelena <email@hidden>
- Date: Tue, 13 Mar 2001 23:19:59 -0500
>
How far should I go to optimize the linearization target? Instead of No
>
Color Adjustment in the driver, Color Controls without adjustments gives me
>
a more even tonal distribution of their linearization target, albeit with a
>
lot of blue, and a bit bright. Why not adjust the sliders a bit? I could
>
make a print of an image look as best as I can with those controls - keeping
>
Same as Source in the PS6 pull down menu, and using those settings on my
>
linearization target, and saving those settings for subsequent use with my
>
new profile.
That is definitively a good idea. As long as your choice is not
"automatic" (which as far as I know means variable), optimizing the
driver results before profiling should result in an improvement.
>
Furthermore, if that would work - would it? logic tells my feeble mind it
>
would - why not double profile for that matter? Why not select a profile in
>
the PS drop down menu that gets me close, use it to print my lin. target and
>
profile building target. Then when my new profile arrives keep that original
>
profile selected in PS, and apply the new custom profile in the Colorsync
>
section of the driver.
Now you're venturing into muddier waters. There are limitations to
instrument accuracy, limitations in the theoretical model, and
limitations in the math routines in the CMMs and in the profile
generators, besides the issue of the profile being a discrete "grid" of
points with a fixed bit depth. If you profile the profile the profile the
profile <grin>, you multiply your chances of getting it wrong.
Scitex ProfileWizard has an "iteration" feature. You build a profile,
then measure the printed output from the profile, and it will compare
that with the Lab values you were supposed to achieve and automatically
"tweaks" the profile to get closer.
They themselves recommend against using that under 2dE, since iterating
further can result in worse results than you already have.
Pixelis has a very interesting color calibration method; they in fact
double-profile, but once in RGB and once in CMYK. So you end up driving
an RGB printer (because they drive through the QuickDraw driver) as if it
were CMYK. Something like this could be tried manually as a means of
getting better profiles, since I am totally convinced that most of the
profiling software manufacturers have invested much more time in building
better routines for CMYK than for RGB.
As for editing, more than once I've found profile editing programs ruin
the black generation or the ink limits. So my personal favorite method
for editing is editing the instrument values before building the profile.
The visual way to do this, very unorthodox, has been to make a TIFF image
of the Lab readings of the original chart (via Logo ColorLab). Then open
it in Photoshop, and place some Lab photos at the side. Then edit in
Photoshop, slightly, to get desired visual appearance (calibrated monitor
needed). Then crop the edited testchart, downsample it so that each patch
is only one pixel, and save as RAW. A linux shell line converts that raw
into tab delimited ascii values, which go to excel, and are saved into a
testchart ascii file.
That one is imported into PrintOpen, and the profile made. That way I
have a first-generation profile, concordant with PrintOpen's internal
model, but in fact edited.
best regards,
-- Roberto Michelena
EOS S.A.
Lima, Peru