Hi,

> I've seen, that Marti has already implemented an
> optimization for RGB -> RGB transforms, which could be
> probably extended for XYZ. Marti?

Yep, that is. I was thinking on same when your mail just arrived ;-)
XYZ is a sort of RGB which goes never on negative values, so it
probably would be prone for pre/post tables.

The actual problem happens only with precalculated transforms
from XYZ to  RGB, CMYK or whatever device space. Since XYZ
operates on gamma 1.0 and has huge gamut, the resulting devicelink
CLUT  restricts output gamut to very  few input gridpoints, and as a result,
contouring artifacts arises. This does not happen when converting from,
say CMYK to CMYK because gamuts of input and output are alike,
not same but no too different. When differences of gamut are
huge AND the biggest gamut space is used as input, CLUT does not
work so well. Lab has either big gamut, but since it is perceptually uniform
the problem is minimized.

That was, indeed commented in appendix B of tutorial.

Regards,
Marti.

----- Original Message ----- 
From: "Gerhard Fuernkranz" <[EMAIL PROTECTED]>
To: "Kai-Uwe Behrmann" <[EMAIL PROTECTED]>
Cc: <[EMAIL PROTECTED]>; <[EMAIL PROTECTED]>
Sent: Monday, September 20, 2004 12:43 PM
Subject: Re: [Lcms-user] over white color clipping



> I tested lcms-internal and -external XYZ profiles. There is
> an quality drawback with XYZ with lcms default settings.

Hi Kai-Uwe,

I think the problem is that the precalculated device link uses only a
n-dimensional CLUT table, but no input and output tables. However, finding
optimal input and output tables (which e.g. make the CLUT grid as flat as
possible, to minimize interpolation error) for the general case, i.e. for an
arbitrarry profile chain is computationally expensive (you basically need to
perform numerical optimization), and I guess, you're not happy either, if
the accuracy is good, but if cmsCreateTransform() takes several minutes.
Such an expensive computation would be acceptable for profile creation, but
not at runtime, when a profile gets applied. I think, the only reasonable
trade-off might be probably to use some heuristics for creating input and
output tables for the device link. I've seen, that Marti has already
implemented an optimization for RGB -> RGB transforms, which could be
probably extended for XYZ. Marti?

Regards,
Gerhard

-- 
Superg�nstige DSL-Tarife + WLAN-Router f�r 0,- EUR*
Jetzt zu GMX wechseln und sparen http://www.gmx.net/de/go/dsl



-------------------------------------------------------
This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170
Project Admins to receive an Apple iPod Mini FREE for your judgement on
who ports your project to Linux PPC the best. Sponsored by IBM.
Deadline: Sept. 24. Go here: http://sf.net/ppc_contest.php
_______________________________________________
Lcms-user mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/lcms-user

Reply via email to