Numerical Analysis on: Photoshop Euro v2 profile & troubles
Numerical Analysis on: Photoshop Euro v2 profile & troubles
- Subject: Numerical Analysis on: Photoshop Euro v2 profile & troubles
- From: Roberto Michelena <email@hidden>
- Date: Wed, 10 Dec 2003 21:05:54 -0500
I had always had the impression that Adobe's Euroscale v2 profiles were very
good. And thus, reading through this thread full of unqualified statements
such as "has a cyan correction", "lacks grey balance", etc surprised me a
lot.
So I've done a little numerical analysis (with a procedure I developed some
time ago), and it demonstrates that Adobe's profiles adhere to the Fogra
dataset from which they were built, EVEN BETTER than Gretag's profiles (that
as far as I know are the same as the ECI profiles), in their colorimetric
tables. In perceptual, it's another story.
Disclaimer: I used what I had at hand while I was offline; the old COMMSP
datasets published by Fogra in the year 2000 (or 1999?), from which Adobe
developed its profiles and which, to my knowledge, are still the official
ones. ECI published, from what I remember, alternatives with "bb" and "sb"
versions (black backing and substrate backing) but the COMMSP datasets were
all black backing as specified by ISO.
And also, the profiles from Gretag's website (dated 2003) that have bb and
sb versions, and that should be equal or better than the ones Fogra
published at some time that were also built with ProfileMaker but an older
version.
tools: Gretag's ColorLab and MeasureTool
****************************
First, I compare each profile's fidelity to the original FOGRA data for
sheetfed coated, a text file called COMMSP1 (from IT8.7/3)
For each profile, I pass the CMYK IT8.7/3 ref file through its device->pcs
table, in absolute colorimetric. I then compare this abscol Lab values with
the original Lab values in COMMSP1
Adobe's Euroscale Coated:
avg dE = 2.77
best 90% = 2.49
worst 10% = 5.35
max = 7.78
Gretag's ISO Coated 150lpi black backing:
avg dE = 4.18
best 90% = 3.69
worst 10% = 8.57
max = 14.08
I chose Gretag's "bb" profile since it yields better results than the "sb"
profile (understandably, since COMMSP1 is black backed).
You can clearly see that Adobe's "device->pcs" table (the part of the
profile used for proofing) is closer to the Fogra data than Gretag's.
****************************
Second, for each profile I test it's roundtrip integrity; this is
accomplished by the following procedure:
1) select a small-gamut set of Lab patches that will always be in-gamut for
the tested profiles; I create one passing IT8.7/3 CMYK ref through a
newsprint profile (Quiz from Ifra's website) in abscol mode.
2) taking that Lab file, convert to CMYK in abscol through the tested
profile
3) taking that CMYK file, convert back to Lab in abscol through the tested
profile
4) compare the resulting Lab to the original one
This yields an indication of how much the forward and reverse tables of the
profile are matched or in agreement. If you tweak the pcs->device table
(separation table), for example, you lose this agreement.
Adobe's Euroscale Coated:
avg dE = 0.84
best 90% = 0.8
worst 10% = 1.18
max = 1.36
Gretag's ISO Coated 150lpi black backing:
avg dE = 0.50
best 90% = 0.47
worst 10% = 0.81
max = 1.09
Both are very good. Gretag is a tad better, but that is probably because
it's device->pcs table is not as tightly fit to the original data as
Adobe's, thus it has room to be fitter to it's own algorithms. Which is not
necessarily a good idea, but will look good on screen.
****************************
And finally, for curiosity's sake, I tested how much did the perceptual
table in each profile deviates from colorimetrical (to see how tweaked it
is). For this, I use the same "roundtrip" procedure outlined above, but the
Lab->CMYK done in perceptual (the CMYK->Lab remains absolute).
Adobe's Euroscale Coated:
avg dE = 4.10
best 90% = 3.63
worst 10% = 8.32
max = 13.22
Gretag's ISO Coated 150lpi black backing:
avg dE = 1.73
best 90% = 1.49
worst 10% = 3.86
max = 4.75
What does this mean? Basically that Gretag's Perceptual table is more
"colorimetric" and less "perceptual"; of course related to this "newsprint"
set of colors that are all well inside the gamut. It should become
"perceptualer" towards the edge.
So I would think Adobe's perceptual is probably a better choice when
separating from a large gamut image (for example those Fuji Velvia
transparencies), because it is smoothing out the gamut compression over all
the range. While Gretag's is likely to be a better choice when smaller gamut
images are separated. But in that case, an ever better choice according to
what I've read, is to use colorimetric instead of perceptual, color
correcting the out-of-gamut colors before the conversion, to force them into
gamut.
But something like "heavy cast" or "gray unbalance"... The numerical results
don't back those claims at all for the colorimetric intents, which should be
the choice for most professional color separators.
About the perceptual intent... Could be; but when building a perceptual
table compromises and choices have to be made, according to the users or
conditions one is targeting. I guess Adobe targeted AdobeRGB and general
inexperienced users? It will be a very nice day when we get some kind of
"image specific perceptual", but it's a concept whose workability still has
to be, er, worked out.
Ah! and I also agree that Adobe should take Fogra's ISO Web Offset data and
build and bundle an "European Web Offset" profile the same way that they did
with the sheetfed one. THAT is indeed something missing.
One last thing: Martin Orpen and Daniel Metz both say that ISO and Adobe
profiles look different (shifted colors) in proofs and in PrintOpen's view.
However the absolute colorimetric intent yields very similar results
numerically (through Apple's CMM of OSX 10.3), with Adobe's closer to Fogra
data. So maybe whatever proofing software in use is having trouble with
white point adaptations or with the profile's gridpoints, we have seen that
in the past very frequently...
-- Roberto Michelena
EOS S.A.
Lima, Peru
_______________________________________________
colorsync-users mailing list | email@hidden
Help/Unsubscribe/Archives:
http://www.lists.apple.com/mailman/listinfo/colorsync-users
Do not post admin requests to the list. They will be ignored.