[argyllcms] Re: Creating a ICC profile for Camera using custom reference target
- From: Ben Goren <ben@xxxxxxxxxxxxxxxx>
- To: argyllcms@xxxxxxxxxxxxx
- Date: Thu, 21 Dec 2017 16:07:24 -0700
So...a number of points.
On Dec 21, 2017, at 2:19 PM, Willemijn Elkhuizen <W.S.Elkhuizen@xxxxxxxxxx>
wrote:
'Artist Paint Target', which is developed by prof. Roy Berns,
Dr. Berns is probably today's leading expert on art reproduction. That writ...
which should be more suitable for painting reproduction than for instance the
ColorChecker. According to his research specific pigments used in paintings
should be better distinguishable with this target.
...while I'd immediately accept that he's probably created the best-to-date
commercial target, even he would agree that target-based profiling is at the
bottom of the ladder when it comes to reproduction. If your studio setup is
perfect, you can get good results...but, if you're going to go to the effort of
getting your studio setup perfect (which is an absolute-must starting point),
the effort in doing something better than target-based profiling isn't all that
big a leap.
Ideally, you want to take one of the multi-spectral approaches that Dr. Berns
and his lab has developed. There are many. Equipment investment ranges from the
insanely expensive to ridiculously cheap. An high-resolution spectroradiometer
might cost more than your house. At the other end, a couple well-chosen color
(Wratten) filters over your lens could effectively get you to ~50nm spectral
resolution, which is probably as much as anybody needs in the real world.
But any of those will take you far off the beaten path. Setting up such a
studio might be comparable to a masters-level thesis project.
If that's a bridge too far, especially if you want to stick with standard RGB
file formats and similar familiar tools, you can get superlative results for
commercial reproduction purposes by generating your ICC profiles from a
spectral model of your camera's sensor. Basically, you determine what RGB
values your camera will output for any given incipient spectra, and use that
data to build your profile. The brute-force method involves mathematically
generating a massive .ti3 file from simulated data and feeding that to Argyll's
regular profiling tools.
Such an approach will also incorporate, for example, the
(spectrometer-measured) illuminant you're using in your studio. Whatever your
illuminant is, it's not D50; however, the way the profiling math works out, if
you do it right, the end result is as if your studio lighting really were D50.
See Anders Torger's DCamProf for a set of tools that can get you started on
this path...but you'll probably have to build a bunch of tools for yourself as
well.
the colprof function just fails at the very beginning, saying that I have too
many data rows (25 or 26 instead of 24), even though this is definitely not
the case.
The file formats aren't novice-friendly. You really do have an error in the
file somewhere. When I was getting started, it took me some time to spot
similar problems. Check for things like extra linefeeds, too many columns in a
single row, that sort of thing.
I could also enter the spectral reference data from the target, this is also
provided in the reference document. Would this make a difference for the end
result?
Argyll will use the spectral data if you specify the illuminant, and in other
circumstances. For what you're doing, you probably want the spectral data.
Another question: if I were to combine reference and measured values from the
Colorchecker DS with the custom target into one .ti3 file, should this (in
theory) lead to a better result?
More samples are almost always better, with the caveat of GIGO. Photographing
charts is particularly difficult, and the errors from imperfect technique can
multiply in unexpected ways from different shots of different targets.
...but I'll also note that your shots of the artwork will manifest similar
errors. It doesn't take much to introduce surprisingly-large errors from glare,
reflections, spill, uneven illumination, and much more....
b&
Other related posts:
