[argyllcms] Re: Posterization caused by profiles

  • From: Graeme Gill <graeme@xxxxxxxxxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Mon, 11 Dec 2006 14:51:45 +1100

william.wood@xxxxxxx wrote:
I'm not familiar with specifying the gamut mapping.  Are you saying I
should specify that the profile is to be used only with, say, images that
are in the sRGB space?

Yes. By default, profile will generate a profile with just a
colorimetric table, giving absolute or relative colorimetric
intent. Only if a source gamut (usually defined
by a profile) is supplied can perceptual or saturation
intent tables be generated (You can't really do gamut
mapping if the source gamut is unknown). To avoid
gamut clipping, perceptual or saturation intent needs
to be used. For the best result (ie, say printing), then
I really recommend "smart" or "active" linking be used
(icclink -G), resulting in a device link profile. This
allows more flexibility in choosing intent, and a far
smoother transform than using the B2A table in a profile,
even if it is configured for the source profile in question.

I've never had to do this with Monaco OptixPro software, and it doesn't
produce the posterization so I was surprised to see posterization in
Argyll.  I tried creating the profile again with -qh and it improved
somewhat; I'm tempted to try -qu now ;-)

Sorry, it's hard to draw conclusions without any details. Other
profilers fudge perceptual gamut mapping, often using a generalized
gamut mapping. Sometimes they give you some controls to tweak it.
This may well be smoother than gamut clipping, but will be far from
optimal for a given transform.

profile -v -A "Lenovo" -M "T60" -D "T60 IPS" T42pk1f500

  What intent tables were built.
Don't know - can you tell by the commad line?

Yes, you only built colorimetric intent.
See <http://www.argyllcms.com/doc/profile.html#s> and
the discussion in <http://www.argyllcms.com/doc/Scenarios.html#PP5>.

Don't know - must you specifiy this?  I thought that was an optimization
only, that you could still get a good profile without limiting it to one
source profile?

It depends on what you mean by "good profile". To some people,
a good profile is one that represents the behaviour of the
device well, ie. how good it's colorimetric A2B table is,
since this is what captures the essential character of the
device, and with a "smart" linker is what counts when using
it for output, or for recreating B2A tables.

If you intent to use it in conventional "dumb" linking scenarios,
then the characteristics of the various intent B2A tables (which
are the ICC way of trading speed for accuracy) will be of interest
to you.


Well you should (I hope!) get a better result if there was a perceptual
table in the profile, especially if it was created for an sRGB source gamut.

Using MS or Adobe color engine - are they passive?  I don't know for sure
what the difference is.

In general it's passive ("dumb"), the only exception being the BPC
feature, which actively alters the gamut mapping in the luminance axis.
You shouldn't need/use BPC for an Argyll profile perceptual intent that
has been created for the particular source profile you are using though.

So try again using something like:

profile -v -A "Lenovo" -M "T60" -D "T60 IPS" -S srgb.icm T42pk1f500

and see if it is different, and possibly looks somewhat better.

Graeme Gill.

Other related posts: