[argyllcms] Re: ColProf considerations...
- From: Ben Goren <ben@xxxxxxxxxxxxxxxx>
- To: argyllcms@xxxxxxxxxxxxx
- Date: Fri, 11 Mar 2016 08:23:55 -0700
On Mar 11, 2016, at 5:53 AM, Massimo Colagrande <massimo@xxxxxxxxxxxxxxx> wrote:
• is there any available color chart with thousands of patches (not
only 144 as in the CCSG), with a large gamut that could be used and maybe
provide more accurate results on a wider set of colors ?
Yes...and no. It's complicated.
If you can create a model of the scanner's RGB response to arbitrary spectra,
you can create a virtual color chart with as many patches as you like. There're
challenges in doing that even with a camera, but let's handwave that away for a
moment and assume you can do so.
The closer you get to the spectral locus (the more saturated the colors), the
ability to profile the device depends more and more on how closely the device's
RGB response matches (at least mathematically) the spectral response of the
human eye. In practice, modern devices differentiate between spectra that look
identical to humans, and, much worse, _don't_ differentiate between spectra
that do look different to us. And you get all sorts of weird inversions and
loops and the like that can make mathematical hash of trying to make a map
between the one and the other.
So...I know very well that the first instinct is for an all-encompassing
profile that gets the entire gamut out to monochromatic colors, but the
practical reality is that you're not going to get that...so you have to decide
what compromises you can best live with.
For a scanner, I'd assume you're doing much more fine art reproduction than
anything else. If the scanner's response is reasonably linear, you should be
able to do very well hand-crafting your own custom charts using the same media
as the art itself. That's especially the case if you have a working
relationship with the artist who created the work. You'll want a representative
sample of all paints (whatever) used in different tint levels and mixtures,
laid out on a grid. The more patches the better, of course, but you don't need
to go overboard...just fill a letter-sized sheet with squares, at least
inch-sized squares and no smaller than half-inch. Start a collection of such,
and pick the one that matches the art.
If you want to build a virtual model of the scanner...you're on your own.
You'll need to know the spectral sensitivities of the red, green, and blue
channels. If the scanner's response isn't linear, you'll need to know how it
deviates from linearity...possibly by channel in the worst case. And you'll
need the SPD of the illuminant. With a camera, you can get what you need by
photographing the output of a spectroscope combined with intelligent use of a
spectrometer, and you can build a large high-qualiy spectroscope for almost
nothing. For a scanner...no clue what the best approach is to all that.
And don't forget the other half of the pipeline! Once you've got this fantastic
image captured from the scanner...what're you going to do with it? Unless your
printer can reproduce those colors, all you've got is a mathematical fiction.
You might experiment with printing your own chart. Yes, it's not necessarily a
good spectral match for the input...but it's a _perfect_ spectral match for the
output. You might find that, messy as it theoretically is, you get printed
results as good that way as any else.
Cheers,
b&
Other related posts:
