[argyllcms] Re: Perceptual intent

  • From: Graeme Gill <graeme@xxxxxxxxxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Tue, 16 Dec 2014 22:42:03 +1100

robert@xxxxxxxxxxxxxxxxxx wrote:

> For example, here is a comparison between a paper profiled with i1Profiler
> and one profiled with Argyll:
> i1Profiler Colorimetric:  DMax 2.44, Volume: 93,000
> Argyll Colorimetric:      DMax 2.5,  Volume: 97,000
> i1Profiler Perceptual:    DMax 2.46, Volume: 92,000
> Argyll Perceptual:        DMax 2.08, Volume: 90,000

Hi Robert,
        you don't say how you are computing these numbers.
When I look at a profile, I typically measure it's gamut
from the absolute colorimetric data since that represents
the underlying device limit, so it's not clear how you are measuring
a "Perceptual Gamut".

Similarly for DMax - profiles don't contain density information,
since it uses completely differet CMF's to the CIE observer.
L* value of black is a more typical characterization,
and, it's not clear how you are measuring a "DMax".

> This is from Beta RGB, with Beta RGB specified in colprof (i1Profiler
> doesn't require a source profile to be specified).  As you can see, the
> colorimetric mapping is excellent in Argyll, whereas the perceptual mapping
> is not at all good.

Beta RGB is a large gamut space. Unless you have an image that completely
fills this gamut, you would never use it as a source gamut for a print
conversion - you will get excessive compression, and very dull
looking output for typical images. If your images are stored in a large
gamut space, then you really have to use a source gamut that better represents
the gamut of what you are mapping, either a more typical output referred
space such as sRGB, or a gamut created from the specific images.

> It seems as though your interpretation of Perceptual is very strict, with an
> overall squeezing in of the out-of-gamut colors.  This results in a lower
> gamut volume ... and, more critically, a very much lower DMax.

I don't see anything like that for typical RGB source spaces. Typically
the RGB space is idealized and has a zero black, so the output it
the minimum value possible in the print space, consistent with the black
point neutral axis constraints.

For instance, lets take Fogra39 with a 300% ink limit:

 xicclu -fif -ir -l300 Fogra39.icm
  0.000000 0.000000 0.000000 [Lab] -> Lut -> 0.934413 0.764857 0.275418 
1.000000 [CMYK]
 xicclu -ff -ir -l300 Fogra39.icm
  0.934413 0.764857 0.275418 1.000000 [CMYK] -> Lut -> 9.068751 1.323905 
-3.212585 [Lab]

 collink -v -ir -G -l300 sRGB.icm Fogra39.icm test.icm

 xicclu test.icm
  0.000000 0.000000 0.000000 [RGB] -> Lut -> 0.746010 0.642229 0.507217 
1.000000 [CMYK]

 xicclu -ff -ir -l300 Fogra39.icm
  0.746010 0.642229 0.507217 1.000000 [CMYK] -> Lut -> 10.606275 0.116586 
-0.063567 [Lab]

So L* of 10.6 vs. 9, but the gamut mapped black is much more neutral.

> If this is not a bug and not a user error, then, IMO, the perceptual mapping
> in Argyll is almost unusable as is.

That's not the feedback I get from most people - it's generally regarded as very
good, often better than the alternatives.

Graeme Gill.

Other related posts: