Fwd: Ugra/Fogra Media Wedge v3.0 [was: Colorsync-users Digest, Vol 5, Issue 299]
Fwd: Ugra/Fogra Media Wedge v3.0 [was: Colorsync-users Digest, Vol 5, Issue 299]
- Subject: Fwd: Ugra/Fogra Media Wedge v3.0 [was: Colorsync-users Digest, Vol 5, Issue 299]
- From: Mike Strickler <email@hidden>
- Date: Wed, 10 Sep 2008 23:05:59 -0700
Begin forwarded message:
From: Mike Strickler <email@hidden>
Date: September 10, 2008 10:49:17 PM PDT
To: Klaus Karcher <email@hidden>
Subject: Re: Ugra/Fogra Media Wedge v3.0 [was: Colorsync-users
Digest, Vol 5, Issue 299]
Klaus,
This exchange is taking quite a lot of time--much more than I would
need to evaluate the problem and repair it, which I could only do
if I had more evidence than what you've given me. For me this all
comes down to two things (perhaps...): Martin's alarm at seeing a
dE of 4.4 on an extremely dark patch and his word (according to
your account) that shadow detail is lacking in actual images. Maybe
true maybe not--let's assume it is. I can tell you that I have
never had this problem with PM5 on either an Epson or HP Z-Series
printer. I don't know what you mean by "shadows" when you speak of
dEs of 3.7/6.7--this may or may not be noticeable. I can tell you
that no further discussion of it will clarify it without looking at
actual prints, which unfortunately is also a paid activity :-).
But to dispel the notion that one cannot get good results with PM5,
I can tell you that I regularly achieve max dE on a full IT8.7/4
against a GRACoL reference (very close to Fogra 39) for all patches
(including the worst 5%) of less than 5, sometimes less than 4,
using PM5 and a few iterations, with avg. dE of all patches in the .
6 range and grey hue error of down around .2 or so. This is just
routine, as are proofs that have smooth gradations and other
"immeasurables." So I just have to throw up my hands and say either
there really was no problem, or if there was I don't know what
caused it and wouldn't know unless I saw it myself.
I know there is a certain appeal to the arcana of color
measurement, and I'm glad someone is doing this or we wouldn't have
such wonderful tools, but for me and my customers it comes down to
the proof itself: if good numbers are confirmed by a good visual
match everyone is happy, and we have been. If you and Martin
haven't been, one would have to consider the visual evidence as
well as the numbers and start investigating. But I reject
categorically that Profilemaker cannot make splendid inkjet
profiles, Martin's reported bad luck notwithstanding. I'm afraid I
must move on and let you carry on with this research. Keep us posted.
Best regards,
Mike
On Sep 10, 2008, at 7:19 PM, Klaus Karcher wrote:
Mike Strickler wrote:
Thanks, Klaus. Not sure if I understand the context anymore. [...]
Has Martin made proofs with these different profiles, iterated
them all, and found some proofs lacking in their images as
opposed to a single number on his verification label? Let's keep
this real...
Yes, he did that (at least with two of them) and told me that he can
clearly see the lack of shadow detail and contrast even on the
testcharts.
I want to refer to Martin's initial question:
If you are, what sort of results are you getting for patch A22 --
it's ruining my approval stickers by scoring 4.4dE on ISO Coated
v2 :-(
As the acceptable tolerance for the max. Delta E on the Fogra
MediaWedge
is 6, Martin already uses up 73% of the tolerance range. I don't
think
that this is a good starting point for a newly equipped proofing
system:
even a small drift of the system or the usual differences between
instruments can render his proof invalid as contract proof
according to
the German MediaStandard print. It's not the nasty number on the
approval sticker one should worry about but the question whether
or not
a referee would accept it as a proof in a law suit.
I agree with you that one should not overestimate single outliers,
especially on the dark end of the gamut -- therefore I wrote:
BTW: large Delta E's at the dark end are not pretty, but I think
one should not overestimate them as colors with L < approx. 12
are rare in practice. Furthermore I think FOGRA39 is a very
"optimistic" definition of the capabilities of a offset press in
this regard: I guess one will not run across offset prints on
matte coated papers with Lmin about 8 very frequently in
practice. Am I right?
But I suspected that Martin's outlier was only the tip of the
iceberg. I
guessed it was a gamut issue and was right: He measured my test chart
and it revealed a mean/max Delta E of 3.7/6.7 in the shadows.
The reason for the bad result was the profile that shortened the
device gamut -- and even the 100st iteration can't bring it back. The
Argyll profile made from the very same measurement did not shorten
the gamut. Martin wrote:
... the shadow patches in column 22 are all down below 1.5dE instead
of the 4.4 to 4.7dE that I was getting with the original profile!
... and this is exactly what one can also see in the gamut
screenshots I
posted.
you wrote:
From these graphs I cannot tell what L* value this "shortcoming"
is at.
But you see the /reason/ for the shortcoming, don't you? The white
body
in my screenshots is the ISO Coated v2 gamut and the missing piece
is --
according to Martin's measurements -- up to Delta E 6.7 thick.
you wrote:
What we need in proofing is the ability to match a reference, not
profile quality in the abstract
I can't fully agree: what we need is a good *visual* match between
the
proof and the (reference-)print. Some of the visual properties (like
accuracy at several points of the colorspace) are easy to measure --
others (like the smoothness of gradients) not.
The iteration techniques exist solely to improve this match.
They exist to improve the accuracy on certain points of the
colorspace
-- but when they are implemented sloopy and apply corrections too
locally, they still improve the numerical accuracy when one tests
them
against the training set, but they corrupt the smoothness of the
profile
and therefore the visual quality of the proof.
Separating data into a training set and an independent test set is
common practice and indispensable in many scientific fields --
amongst
others when testing regression models. And profiling is nothing
else but
applying regression models and interpolation algorithms. One still
can
not measure properties like smoothness this way, but at least one can
unmask that local "cheating" does not improve the overall quality
of a
profile.
Klaus Karcher
--
Klaus Karcher * Eichenallee 18
26203 Wardenburg * Germany
Tel. +49 441 8859770
_______________________________________________
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