Re: Media Testing for maclife.de
Re: Media Testing for maclife.de
- Subject: Re: Media Testing for maclife.de
- From: Uli Zappe <email@hidden>
- Date: Sat, 13 Sep 2008 23:06:32 +0200
Am 13.09.2008 um 14:11 schrieb Eric Chan:
Calibration would bring the data from the device into a standard
No, that's the task for the profile (or at least, should be in a
clean workflow).
I disagree on this point, Uli. In general, for most devices
calibration data is both stored and applied separately from actual
profile data.
This was simply a misunderstanding. You use "standard" in the meaning
of "well-defined" state, and I fully agree that's what calibration
(and not profiling) is for. With "standard" I associated something
that is publicly defined, such as a standard color space (e.g. Adobe
1998), but maybe Chris used the word in the narrower meaning you did.
For example, for Epson large format printers there is calibration
data generated and applied using Epson ColorBase software, which is
separate and independent of ICC profile data. The idea being that if
two printers of the same model are calibrated using ColorBase
(effectively linearizing each channel), then you should be able to
share a single ICC profile between them with equally good results.
The ColorBase concept has confused me to some extent. It tries to take
the calibration step from "well-defined for a single device" to "well-
defined across all devices of one type", thereby enabling the use of
generic ICC profiles you won't have to measure and create individually
anymore. But what's gained with that step if now you'll need the
spectro for the ColorBase measurements? The only difference
(advantage, depending on your POV) is that the "recipes" for handling
perceptual rendering etc. are now in the hand of Epson and not in the
hand of some manufacturer of profiling software.
It has been suggested that for digital cameras, the appropriate
analogy would be white balance, which (most of the time) is just
effectively scaling the individual channels in camera-native space
before applying the profile data.
Exactly.
That is only part of the story. If there is reference gain data for
the raw file that varies from unit to unit within a given model,
then this reference gain data should also be applied as a
"calibration step" -- separately and independently from any (ICC)
profile data that is later applied.
Again, complete agreement. Each device should neutralize its response
as much as technically feasible during the calibration step.
Any measurement/profiling that takes place without such calibration
being applied is effectively "burning in" a particular unit's
calibration into the measurements & resulting profiles.
Which is pretty much the normal situation, since calibration usually
is not good enough to generate a well-defined state *across different*
devices. Although a monitor is always calibrated before it's profiled,
usually you would not consider exchanging the profiles of two monitors
(of the same type).
Going back to the printer analogy: if you didn't linearize the
printer first (e.g., run Epson ColorBase) and just proceeded with
printing a standard profile target and building a profile (e.g.,
using your favorite profile-building software), then that profile
should work well on that unit. Any peculiarites of that particular
printer unit are effectively "burned into" the profile that you've
built. But if that unit behaves differently from another unit of the
same model, you would not expect the profile you built for the 1st
unit to necessarily work well for the 2nd.
Hope this is clear.
It's perfectly clear, but it evokes an interesting question I wasn't
able to answer convincingly so far.
Because ICC camera profiling software needs a TIFF target image, the
rendering from RAW to TIFF, including setting the white balance, will
always have been done before profiling. And since setting the white
balance can move the color temperature so enormously (contrary to a
scanner, where the scanner only has to compensate for subtle color
shifts of the scanner lamp), the profiling software really cannot
compensate for any calibration that's been "less than perfect", at
least if the profile is meant to be used for different lighting
conditions. This differs from profiling other devices where the
profiling software can do that (although, as you correctly point out,
the profile is then valid only for this specific device, which, again,
is the normal case today.)
So, following from this, you would think that all that ICC camera
profiling software can possibly do is to create "generic" camera
profiles, in the meaning in which an EPSON provided profile for
ColorBase calibrated printers of one type would be generic. But if
this the case, then ICC camera profiling software should not be able
to improve on existing such generic camera profiles as long as those
were built with the goal of delivering metrologically exact results.
However, my measurements clearly show that this is not the case (i.e.
there is a clear improvement when using custom built ICC profiles).
The only two reasons I can think of are these (but maybe there are
more, and I would be eager to learn about them):
1.) Not a single available generic camera profile out there (for a
given camera type) was built with the goal of delivering
metrologically exact results.
2.) There are deviations of color reproduction in camera production
charges that induce more of a "general color cast" than specific
spectral deficiencies, and thus can be compensated for by ICC
profiling software even after the white balance was applied, whereas a
generic profile cannot take them into account.
If 2) is the case, then fine, but I'm unsure whether this is plausible
technically, and looking at the factual distribution of color
deviation across the color patches I measured suggests the problem on
which ICC camera profiles improve are individual colors, not a general
color cast.
If 1) is the case, this would mean an almost grotesque waste of
resources, because it would mean that thousands of people buy
expensive ICC camera profiling packages and spend hours building
custom profiles, although the simple existence of a metrologically
correct generic profile for their camera would make that completely
superfluous. However, apart from whatever remains of my belief in the
rationality of a free market, the Canon "Faithful" profile also argues
against this variant, because Canon explicitly designed this to be
metrologically correct, and yet a custom built ICC profile clearly
improves on it.
So why do ICC camera profiles work so well as they - for me, without a
doubt - do? Because of 1), 2) or ??
One final remark: in a reply to Edmund, you wrote:
The feedback that we have gotten from most of our users (not color
scientists, enthusiasts, etc.) is that they don't want accurate
scene-referred color.
So, fair enough, your foremost goal will be delivering pleasing
"looks". However, now that you started to actually offer more than one
profile for each supported camera in Lightroom (namely, the Adobe
profile variants plus the emulation of the camera vendor profiles),
couldn't you just add one more alternative and offer a metrologically
correct camera profile as an option for those who prefer that? From my
POV, this would give Lightroom a very real advantage over e.g.
Aperture. OTOH, at exactly this point, Edmund's argument might well be
correct: as soon as Adobe would explicitly offer a "metrologically
correct" profile as an option, this profile - and thus "Adobe" - would
become subject to an easy evaluation by numbers. So maybe that's the
marketing decision behind not even offering such a profile as a mere
option?
Bye
Uli
________________________________________________________
Uli Zappe, Solmsstraße 5, D-65189 Wiesbaden, Germany
http://www.ritual.org
Fon: +49-700-ULIZAPPE
Fax: +49-700-ZAPPEFAX
________________________________________________________
_______________________________________________
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