Re: maclife.de
Re: maclife.de
- Subject: Re: maclife.de
- From: Uli Zappe <email@hidden>
- Date: Mon, 18 Aug 2008 14:56:56 +0200
Am 16.08.2008 um 03:45 schrieb Graeme Gill:
Uli Zappe wrote:
Just FYI, SilverFast, OTOH, does not expand luminosity at all in
the perceptual intent (in fact, there simply is no different
perceptual intent in these profiles), making the images always look
dim, but they never clip. So SilverFast was also deducted 0.5
points for the opposite reason, or else it would have been on par
with the Logo profiles. The Logo profiles seem to hit the sweet
spot in between.
You can choose whether to create an "always absolute intent" profile
using the -u flag in Argyll, rather than the default relative
colorimetric intent table, the former having the benefit of more
L*a*b* PCS headroom for clipping. See <http://www.argyllcms.com/doc/colprof.html#u
>
Yep, but as I wrote, that didn't seem to be ideal, either. Logo's (i1
Match / ProfileMaker) behavior is somewhere in between, and this seems
to be the best.
But I haven't experimented with the -u flag yet (or other flags, for
that matter), and will do when time allows.
Self fit error is not a good measure of ultimate quality, since it
assumes that the data set is a perfect representation of the device
behavior, which it is not, since it is sparsely sampled and contains
measurement uncertainty (noise).
A more rigorous measure would be to use a much higher resolution
test chart (say 5 to 10 x as many test points) as the reference,
removing from it the test points used to create the profile.
Uhm, yes, but that's exactly what I did. (Well, almost 5 times; the
test target actually contained 1152 patches.)
The test set is then an independent test that gives a better measure
of how well the profile represents the underlying device
characteristics. Even this approach has its drawbacks, since the
reference points also contain a lot of noise, making evaluation
indistinct.
Note that (paradoxically) a profile that has a poorer self fit
measure may in fact be a better fit to the underlying device
characteristic, because it has smoothed out some of the noise in the
measurement values, and conversely, a profile that has
a very low self fit error may be a worse match to the underlying
device, since it is slavishly following noise in the measurement
set, and may not be applying a reasonable model in between these
points to generate the interpolated values.
Yes, this is a well known phenomenon not only from scanner profiling
(especially printers come to mind). Profiles with a tight delta E fit
may look quite ugly when applied to a Granger rainbow, for instance,
because smoothness was sacrificed to curve fitting.
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