[Openicc] Re: color-policy vs. color-infrastructure

[Stefan Döhla]

>> Gamut tags are a bit fuzzy at the best of times, and in many profiles
>> out there they are simply wrong. As a result, they can't generally be
>> relied upon.

> Just for the record, it seems Photoshop is not using the gamut tag
> anymore. Lcms is actually using a brain-dead method for inferring gamut
> that works most of times at expense of poor accuracy. The trick is to
> convert values back and forth the colorimetric intent. If color changes,
> then it is outside gamut.

I haven't looked very deeply inside the gamut tags - but I'd like to
mention that there was a very interesting discussion on the german ECI
list about the gamut-warning triangle in Photoshop.
I'm trying to put the explanation of Johannes Hoffstadt in some short
english words:

- the gamut tag is not (known to be) in use - except Apple's ColorSync.
- the PS method works like this:
  a device link  " Lab --B2Ax--> Device --A2Bx-->Lab " is generated
  which is then used to test the shift of the selected Lab value.
  There is a threshold for this shift, which must be exceeded so that
  a Lab value is marked as out of gamut. The device link is generated
  for the currently selected intents.
  This method is applied for
  + RGB, Lab entry and CMYK simulation
  + CMY(K), Lab entry and RGB simulation
  which means, that "device colors" are not gamut-tested

The problem with the PS method is, that it depends on the quality of
correspondence between the A2B and B2A tables. To measure their
quality, easy reproducible colors are probably used for the evaluation.

In PS, the measured quality is not used to set a fixed deltaE as the
allowed difference for the device link, mentioned above. For a general
solution, the threshold should depend on the ab values. This can be
accomplished by using a grid for the Lab space and calculating a
deltaE for every region or more sophisticated methods.

All gamut warning is therefore not exact. It is meant to be a tool
which shows problems that might occur for the selected color.

Stefan

-- 
Stefan Döhla
mailto:color at doehla.de