Re: maclife.de
Re: maclife.de
- Subject: Re: maclife.de
- From: Uli Zappe <email@hidden>
- Date: Tue, 5 Aug 2008 07:16:20 +0200
Part 4, which deals with scanner profiling, has now been published
(again, German only):
http://www.maclife.de/index.php?module=pdfarchive&func=download&pid=2063
The test was performed with a flatbed and a film scanner and
specifically manufactured and hand-measured evaluation targets with
1152 color patches. These targets were from the same production batch
as the (also hand-measured) IT8.7 targets that were used for profile
creation. Thankfully, Wolf Faust (http://targets.coloraid.de),
manufacturer of excellent IT8.7 targets, provided these special test
targets which allowed for much more meaningful test results than
simply testing against the very target that was used for profile
creation (a method that doesn't say too much about profile quality and
delivers far too "good" results).
Here are the summary grades for profile quality (flatbed & film) for
all the tested software. 6 = very good, 1 = unusable, best = first.
ProfileMaker 4.5
i1 Match 4.5
SilverFast IT8 option 4
MonacoProfiler 3.5
EZColor 3.5
basICColor input 3
VueScan Professional 1.5
The most remarkable test result for me was that the profile quality
depends very much on the scanner gamma used. This is true for all
profiling software, but to a different degree. The most extreme
example is the SilverFast IT8 option used with a film scanner: for my
test scanner (a Nikon Super Coolscan 8000ED), deltaE max was ~7 (a
very good result) for gamma 1.8-2.2, but already 10 for 2.3, and 34(!)
for gamma 2.8. Other software produced their best results with
different gammas, so there is no single "optimum" gamma for a scanner.
The results actually differ so much depending on the scanner gamma
that a test with a "random" gamma seems to be almost completely
meaningless. So I ended up building test profiles for all gamma values
from 1.8-2.8 in 0.1 increments, for both scanners with three different
targets each. I then averaged the deltaEmax(gamma) curves for both
scanners separately and used the local minimum as the common gamma for
the respective scanner with which I performed the final tests. In my
case, this was a gamma 2.4 for the flatbed scanner and 2.0 for the
film scanner.
From my results, it is obvious that common recommendations for the
scanner gamma like "always use the gamma of your working space" or
"always use gamma 2.8" are wrong. Ideally, each software package
should provide the user with a validation mechanism that would allow
her to find out the optimum gamma for her scanner/software
combination, but no software does this. (SilverFast and MonacoProfiler
do report quality statistics after profile creation, but both turned
out to be meaningless.)
Other interesting tidbits:
- ProfileMaker/i1 Match and MonacoProfiler/EZColor are both pairs that
provide practically identical results (i1 Match has better error
checking than ProfileMaker, though). This means that for 100 Euro, i1
Match is a clear winner; unfortunately, you need an i1 Pro to be able
to use it.
- Unbelievably, EZColor is voluntarily crippled and cannot use
standard (!) IT8.7/2 targets other than the specific "Monaco" target.
The workaround is to edit the reference file of your IT8.7/2 target,
make the first line read IT8.7/1, and pretend in EZColor you'd like to
profile a film scanner. This works fine, but what a way to treat
paying customers! :-(
- Also, MonacoProfiler and EZColor both use an unreadable, proprietary
format for their reference files. :-( They are capable of reading text
reference files from other manufacturers (EZColor only IT8.7/1), but
these have to be in SimpleText format to be recognized - a format from
the Classic Mac days, dead for 7 years now! One can only hope that X-
Rite gets rid of this ugly Monaco mindset as soon as possible.
- SilverFast's profiles always return absolute colorimetric values, no
matter which rendering intent you choose
- basICColor's profiles have no colorimetric table, so the absolute
colorimetric rendering intent actually returns kind of "absolute
perceptual" values. Therefore, at first it seemed unfair to judge
basICColor input by an evaluation method that uses the absolute
rendering intent to compare produced Lab values to the target
reference, but a careful visual evaluation confirmed that the not-so-
good measurement result for basICColor input corresponded with several
color patches that were not simply reproduced in a "visually pleasing"
manner (=perceptual), but were simply off.
- VueScan is a special case in several regards. It produces matrix
instead of LUT profiles and consequently, cannot use profiles produced
by the other profiling packages. (You can, of course, simply use
VueScan's RAW output and apply a profile yourself, but be careful to
lock exposure time in this case (if offered), or you won't get
reproducible conditions.) VueScan's author, Ed Hamrick, is of the
opinion that it's technically impossible/meaningless to try and
reproduce/map absolute luminosity values, so VueScan does not do this.
As a result, all images with VueScan's profile applied to them look
far too dark unless you apply an auto contrast setting (called "Color
Balance > Neutral") to them which of course maps each image to the RGB
0-255 range *individually*. So every "objective" brightness
information of the scanned images is indeed lost, and not
surprisingly, my measurement results for VueScan were bad (although I
tried hard to correct image brightness and contrast to get optimal
results). Ed Hamrick actually considers comparing Lab values using
absolute colorimetric rendering a "flawed" evaluation method. What
makes matters worse is that due to a VueScan bug, the "Color Balance >
Neutral" setting won't work correctly when applied to 48 bit scans.
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