Re: Camera profiling with ICC et al
Re: Camera profiling with ICC et al
- Subject: Re: Camera profiling with ICC et al
- From: email@hidden
- Date: Mon, 15 Sep 2008 05:01:41 EDT
Uli,
I hoped not to get into this level of detail, but I think this thread shows
some opportunities for improved understanding by many that I'd like to help
with.
1) Uli wrote:
<< Do you think everything that's written on paper is correct just because
it's written there?>>
An ICC white paper represents a consensus of the world's top ICC experts. It
undergoes a peer review process in many ways more demanding than a ISO
standard, US patent, or scientific journal. That doesn't make it perfect, just a
pain-in-the- to get one through. I have been fortunate (I think) to serve is
several of these capacities, so I speak from experience.
2) Uli wrote:
<<I'm a bit at odds with the scene-referred/output-referred distinction for
theoretical reasons>>
Me too. It under appreciates the value of appearance estimation and
preference layering. But that's a different matter. For the test you are running, the
scene-referred/output-referred distinction is absolutely essential. Here's why:
A) All v2 ICC colorimetry is picture-referred and may assumed to be
output-referred. This means that all v2 ICC rendering intents represent the colorimetry
of the desired reproduction, not the scene. Yep, even for the colorimetric
rendering intent.
B) v4 was silent on image state so most assume that v4 also represents
output-referred colorimetry.
C) In v2, v4, and many other non-ICC profiling methods, there is no way that
I knew of to commmunicate weather the colorimetry is scene-referred or
output-referred. This means you have to assume its output-referred, the colorimetry
of the reproduction, not the scene.
D) For this reason, I introduced the "colorimetric intent image state" tag
for v4 profiles. It was approved by the ICC last year and will hopefully start
showing up in camera profiles in the future. If you have a v4 profile with the
image state tag set to one of the scene-referred values, then you know the
profile represents the colorimetry of the scene instead of the desired
reproduction. Otherwise you have to assume the colorimetry is output-referred.
E) It is perfectly understandable that v2, v4, and non-ICC profiles produce
output-referred colorimetry: most workflow demands it. Is only been relatively
recently (RAW) that applications are starting to understand the value of
scene-referred colorimetry. But many RAW workflows also use output-referred
profiles for the sake of productivity.
F) So, in the general case, you have no way of knowing if a given profile
(ICC or otherwise) is intended to produce scene-referred or output-referred
colorimetry.
G) But lets suppose you have a given profile and you know it is
scene-referred, then you are good to go, right? Wrong, for many reasons. Here is just one
example: No shipping camera that I know of meets the Luther-Ives condition. The
means that cameras exhibit significant observer metamerism with respect to
humans. So the scene-referred profile designer must decide how to optimize the
profile. Two possible extremes: i) minimize the color differences of a color
chart, or ii) minimize the color differences of representative scene spectra. If
i) is chosen then the profile will perform optimally on the chart but
suboptimally (not necessarily poorly) on real scenes. Poor choice. If ii) is chosen
then the profile will perform optimally on the scene but suboptimally (and
frequently quite poorly) on color charts. Better choice. If you have any doubt
about this, derive and compare the basis functions for color charts and scene
spectra. ISO 17321 is a good place to start. There is a third possibility
somewhere in the middle. Use a photograph of a color chart to learn something about
the camera colorimetry but influence the profile based on scene-spectra consi
derations. OK choice.
3) Uli wrote:
<<Can you point me to the inadequacy of emulating real-life shots under
a number of different lighting conditions, checking color deviations
of a statistically relevant number of color samples in each of them
and averaging the results of all lighting conditions?>>
Yes. So you capture a target under specified conditions and you apply the
profile, ICC, Adobe, or otherwise. First of all, you don't know if the profile is
scene-referred or output-referred, but most likely the latter, making the
metrology irrelevant. Even if you do know the profile was scene-referred, it most
likely was not optimized to reproduce chart colors even if a chart was used
to help guide the profile building process. Most would prefer that a
scene-referred profile do a good job of reproducing scenes, not charts.
I've used the example of optimization purely for illustrative purposes. But
there are also many other factors that are relevant such as illuminant spectral
power distributions, appearance considerations, and many others. As I said
before, evaluating camera profiles is a very tricky business indeed.
The other Eric
Eric Walowit
Tahoe
**************
Psssst...Have you heard the news? There's a new fashion
blog, plus the latest fall trends and hair styles at StyleList.com.
(http://www.stylelist.com/trends?ncid=aolsty00050000000014)
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Colorsync-users mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden