[argyllcms] Re: Gamut mapping problems in 1.1.0

  • From: qcore@xxxxxxxxxxxxxxxx
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Mon, 09 Nov 2009 00:47:10 +0000

Quoting Klaus Karcher <lists@xxxxxxxxxxxxxxxxx>:

Hi Martin,

you wrote:
I'm using Argyll_V1.1.0_RC1_osx10.4_i86 and the gamut mapping isn't working properly for me when building RGB to CMYK device link profiles.

Everything works as expected until either -g or -G are used. The results are not any better when using source image gamuts either.

There is a loss of saturation immediately noticeable throughout the resulting images.

Images which are highly saturated -- the seaside sunset image in Bill Atkinson's Lab Target for instance -- will not give an acceptable result if any type of gamut mapping is used.

How did you use tiffgamut and collink -g / -G on Bill Atkinson's Lab
images? Did you convert them to RGB? How? Which RGB? Did you cut-out
the individual images before using tiffgamut? (this will make a big
difference!) Did you try tiffgamut -f?

I converted the whole image and sections of the image to DonRGB4 and AdobeRGB1998. My other test image was a Colorport CMYK IT8 target assigned to ISO Coated v2 and then converted to RGB.

My tiffgamut tests were done at -f100 and -f80.

I'm still testing the new gamut mapping but I think there are some
great improvements compared to previous versions. You are right: there
is less saturation than in 1.0.3 or 1.0.4, but there are also much more
details preserved in saturated areas (e.g. in my red flowers example).
While my attempts to improve the gamut mapping parameters in GaMapICC
0.3b caused overdone brightening in some saturated colors, there seems
to happen the opposite in 1.1.0 RC1 (e.g. with roman16_red and ISO
coated v2).

I thought that 1.0.4 already lacked a little saturation and also had a tendency to darken 3/4 tones. Saturated images like the Bill Atkinson sunset/sea image that I mentioned earlier would still be acceptable though even if they weren't as good as the conversion using a non-gamut mapped device link.

But 1.1.0 produces an unacceptable result that is far inferior to an ordinary conversion.

Maybe this issue is related with a separation bug I noticed in
conjunction with collink's -ke flag: I generally get much more black
(and less CMY) than expected when I use collink -ke. There is e.g. no
cyan at all in skin tones, not even in the shadows, where I'd expect
20-30% cyan with a normal ISO Coated v2 separation.

I have also noticed that the cyan channel looks a little harsh in warm colours -- but this happens even if you don't use -ke. For retouching purposes I'm always looking for detail in the weak channels and the strawberry shot in the Bill Atkinson image has no worthwhile cyan detail in the saturated reds -- all the detail is in the black.

Probably haven't noticed this as a problem though because a lot of my separations are to custom versions of the ISO profiles that I've made with Profilemaker to give me lots more black. I still find Argyll "difficult" with regard to black generation in CMYK profiles.

I guess Graeme is on the way to Albquerque [I envy him!], so I think we
have to wait at least one week before he can see about this issue.

OK. Hopefully we can come up with a gamut map wish list before he gets back.

Martin Orpen
Idea Digital Imaging Ltd

Other related posts: