[argyllcms] Re: Question about scanner ICC-profiles
- From: Graeme Gill <graeme@xxxxxxxxxxxxx>
- To: argyllcms@xxxxxxxxxxxxx
- Date: Sat, 20 Feb 2021 11:51:02 +1100
Hermann-Josef Röser wrote:
Hi,
Another observation in this direction is the following. I have IT8-targets from 2 vendors.
One seems to be somewhat underexposed with brightest patches having L = 85, the other has
in the brightest patches L-values of about 93. For the same AG setting, the profile
corresponding to the slightly underexposed target extends not as far in the L-direction as
the better exposed one. I guess the two observations are related to each other?
a "feature" of the ICC profile format is that it stores the basic profiling
information
in a Relative Colorimetric form. This means that the data is normalised to the
white
point. For output profiles this is not of great significance. For input profiles
it has an impact in that it may not model values above the relative white point.
- I'll come back to this point later.
Of even greater impact on making input profiles is the question of gamut. For an
output device you can explore the whole of the possible color gamut by
exploring the whole of the possible device value gamut. General input devices
such as
cameras do not have a gamut in the same sense, since you can adjust exposure to
ensure that the sensors are never saturated, and they will produce a device
output
value for any possible input spectrum. A properly setup scanner should not have
a gamut limit either, putting aside unusual materials such as fluorescence or
specular reflective type materials. (The input devce equivalent of a gamut
limit is really a multi-dimensional plot of spectra vs. colorimetric accuracy.
Not easy to represent, and you need to know the sensor spectral sensitivities to
compute it.)
But a test chart used to characterise an input device does have a limited
gamut. So from a purely information point of view, it's not as accurate to
extrapolated outside the gamut of the test chart as it is to interpolate within
measured data points.
Back to white point - the default ICC representation is white point relative,
and the test chart doesn't test anything above its white point, and the
default ICC data representation is white point relative, so by default
the test chart white comes out at L*a*b* 100, 0, 0.
But viewed in an absolute sense, the L* gamut varies with the test chart white.
The implication is that the input profile is only really accurate when
applied to the same media and process as used to make the test chart.
You can make an Absolute Colorimetric input profile using ArgyllCMS by
using the -ua option. This attempts to make the white point 100%
reflection rather than the chart white. But it has to use extrapolation
to do this, and this can make it inaccurate outside the test chart gamut.
Cheers,
Graeme Gill.
Other related posts:
