[argyllcms] Re: Camera/Scanner Profiling for HDR images
- From: Ben Goren <ben@xxxxxxxxxxxxxxxx>
- To: argyllcms@xxxxxxxxxxxxx
- Date: Wed, 26 Apr 2017 08:28:29 -0700
On Apr 26, 2017, at 1:27 AM, Mike Tough (Redacted sender "mike.tough" for
DMARC) <dmarc-noreply@xxxxxxxxxxxxx> wrote:
I would like to generate an appropriate camera profile to allow me to
maximise the colour gamut of the resulting HDR image.
Um.
First of all, camera profiling involves deep black magic.
The basic problem is that the spectral transmissions of the camera's color
filters differs from the spectral transmissions of the human eye's color
pigments, and it's a mathematical / physical impossibility to get a perfect
match between them.
You know those circles of colored dots with numbers "hidden" in them used to
test for color blindness? You could craft a pair that go both directions -- one
that you'd pass but the camera would fail, and another that the camera would
pass but you'd fail.
The misalignment exists over the entire gamut, but it only becomes
mathematically significant as you approach the spectrum locus. The camera would
do as well as you on the standard colorblindness tests at your
ophthalmologist's office.
And all this is before the fact that cameras are used in _uncontrolled_
environments with all sorts of illuminants. You could craft more colorblindness
tests that you'd pass in daylight but fail in incandescent light, or pass under
one type of fluorescent but fail under another -- and the camera would have the
same magnitude of failures but with different sets of tests.
The practical result is that no profile is going to be perfect, and no
patch-based profiling approach is going to be satisfactory.
If you're prepared to invest a lot of mental resources, you can get excellent
results with a spectral model of the camera's sensor. This is especially true
if the scene's illuminant is known (or can be reasonably approximated). You're
not going to accurately capture highly-saturated colors...but, at the same
time, you're not going to be working with an output device (including
wide-gamut display or printer) that can reproduce them, either.
However, that cost only makes sense if the benefit is comparable. And that
payoff is probably only there for reprographic work, possibly including some
types of commercial photography.
In the case of HDR fine art photography...I would recommend strongly against
attempting to brew your own profiles. If it's artistic effect you're after,
pick the profile you find the most aesthetically pleasing and use it as a
starting point. If you're aiming closer to realism, the pre-canned profiles in
Raw Photo Processor are better than almost any amateur is going to be able to
create -- and more accurate than any of the profiles from the major
manufacturers.
If you really want to start chasing down the rabbit hole, the DCAMPROF project
that Alan mentioned is as bad a place to start as any. That's not to criticize
DCAMPROF...it's more like, "if you really want to wind your own motor for your
own highway-capable electric vehicle, such-and-such drag racing team did it and
put everything online."
But...honestly? This is not a wise investment of your resources....
Cheers,
b&
Other related posts: