[argyllcms] Re: mathematics question - colprof models and cameras
- From: Graeme Gill <graeme@xxxxxxxxxxxxx>
- To: argyllcms@xxxxxxxxxxxxx
- Date: Fri, 16 Dec 2011 12:17:43 +1100
Stephen T wrote:
> So non-linear per channel curve(s) makes the
> channel(s) linearly additive before a matrix transform? In this case, a
> linear RGB file (RAW
> processed with gamma = 1) is a good default input for profiling with options
> -a g, -a G, -a s,
> -a S (and -a x which is linear as well):
Yes, that's the idea. Some people have reported good results by forcing the use
of a simple
matrix (-am) if the device values are linear light, which is often the case for
RAW. You need
a better than 8 bit per component data path of course.
> Fair enough, but in ufraw there is a "linearity" parameter which can be used
> to tune the black
> level and shadows and generally improves profile fit.
Right, but how much of that is compensating for flare ?
> Let's call such modifications of the
> device behaviour "pre-conditioning" the RGB before input to the profile: 1)
> setting gamma = 2.2
> (common for many colourspaces); 2) adjusting exposure to get the white level
> matching the
> reference data (reference XYZ converted to a D50 RGB colourspace, I use Wide
> Gamut RGB); 3)
> adjusting "linearity" to get the black level matching the reference data.
> Excuse me for
> confusing the discussion with an application outside of Argyll CMS!
I'd call that calibration to a typical output type colorspace. You can profile
this of course, and it has the advantage of not requiring so many bits in the
data path, but the converting linear light to non-linear and then hoping the
profiler figures that out and un-does it may be allowing more scope for
errors, or it may make it work better - I don't know, I haven't looked
into it in enough detail.
> "Typically such systems are not setup to be absolute colorimetric, so there
> is likelyto be a
> white point shift, making side by side comparison difficult."
>
> Definitely, but it's important that the profile is consistent with camera
> white balance and
> perceptual rendering intent. I don't want to have to tune the white balance
> on every photo to
> correct slight colour casts resulting from a non-neutral profile. The RAW
> attraction for me is
> white balance.
Hmm. I'm not sure that I understand. One of the issues with attempting to use
profiling
on cameras is that generally the profiling needs to be applied before choosing
the
source white point (white balance). Yet most of the devices in non-raw mode
do the white balance in the camera automatically, before you can apply the
profile.
Raw mode offers the possibility of doing things in the right order, but it means
that you need a workflow that applies the color profile and then chooses a white
balance.
> Well, it looks like I need to do some more comparative evaluations. I really
> don't have enough
> patches for cLUT profiles. Gamma + matrix looks promising.I have also learned
> that the camera
> manufacturer's colour is not always best!
I would imagine that the camera manufacturers (just like the film manufacturers
before them)
are aiming for "pleasing color", not accurate color. So they will fiddle quite
a bit with
the rendering to give the most appealing results. Then there are manufacturing
variations.
Cheers,
Graeme Gill.
Other related posts:
