[argyllcms] Re: Camera matrix profile, adding ti3 perfect white data set

  • From: Graeme Gill <graeme@xxxxxxxxxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Tue, 02 Feb 2010 11:57:51 +1100

Elle Stone wrote:

By "the white and black points will be determined and adjusted after
color management" do you mean setting a manual white balance by
eye-droppering off a neutral target such as a neutral object in the
image and/or a separately exposed white balance card? Or do you mean
using levels to set the brightest bright and darkest dark - changing L
with no regard for a and b values? I'm guessing you mean both.

Any of the above (although it crosses my mind that such methods aren't
the best you can do if they are computed in RGB or L*a*b* space rather than
using something like a Bradford matrix - but that's an application issue).

For a digital camera, white balancing (color balancing) already
corrects chromaticity (right word?) errors in the highlights and
shadows, because the color of the light illuminating the scene
(assuming only one light source and ignoring the fact that objects in
shade are really in a different light source) doesn't vary from the
brightest to the darkest part of the scene. Yes?

Well of course it can, if the darkest object has some reflectance
and it is not neutral (ie. not a flat reflectance spectrum), but it
comes down to how one defines and uses a black point.

But for a printer,
for example, the situation is different. The chromaticity of the
darkest dark can and often is different from the chromaticity of the
white patch (ink vs paper). (And for a monitor?) So for a printer,
"color balancing" so to speak has to be done with respect to both the
brightest and darkest part of the image, along with setting the L

I don't see it as much different, but this is because Argyll doesn't
use the chromaticity of the black point in its perceptual gamut mapping
(other software may do this though). Instead it assumes that the
observers sense of neutral corresponds to the chromaticity of
the media white point, since it is assumed that the image dominates
their view and hence their state of adaptation. So the neutral
axis is assumed to be those colors with the same chromaticity
as the white point. In practice it then "bends" the neutral
axis close to the black point, so as to achieve a neutral
looking result over the majority of the luminance range
with the darkest possible black.

[Note that some sources talk about the observer being
 adapted to the chromaticity of the illuminant, but
 I dispute this is most situations. You can't see an
 illuminant unless it is blinding you, so in practice
 you see it reflected or diffused through other objects,
 and if it has any influence over your state of adaptation
 independent of the image being viewed, then you are in a
 "mixed adaptation" situation, which is outside the scope
 of a simple ICC profile. In practice when looking at images
 it doesn't often matter anyway, since cognitive mechanisms
 complete the adaptation to the neutral of the image.]

One issue with NOT using "-u" is that even the it8 target shot itself
has to be re-white-balanced if it is re-rendered from the raw file and
has the non "-u" profile applied. Otherwise the gs00 patch is rendered
as neutral (R=G=B) and the rest of the image has a yellow cast
(because on my target gs00 actually has a slight blue cast as
indicated by the Lab values in the reference file).

Hmm. It's an assumption in my part that the lightest test patch
is neutral. It's pity the white patch is not neutral in respect
to the other neutral patches for your test chart then. I guess
the white point chromaticity could be computed in some other
fashion that attempts to average the chromaticity of all the neutral
seeming patches, but for typical paper media, the paper
is assumed to be the reference white, and usually people want
the paper background "removed" when they scan documents in
relative colorimetric mode.

Using the "-u" switch in 111beta to make the profile, then applying to
the target during raw rendering, gs00 comes out properly slightly blue
and gs06 comes out neutral, which is what I would expect to happen.
Does this mean I got lucky in my choice of time of day, color of
sunlight when exposing the target? The "-u" profile might not produce
an image already color-balanced to the gs06 patch if the target had
been shot in deep shade? I can see some experimentation is in order.

I think that's the nature of it. The test chart's XYZ reference values
are relative to the illuminant used to measure them, so naturally
if the measurement illuminant is normalized to D50, and D50
is used as the white reference (which it is with ICC PCS),
then a neutral (ie. spectrally flat) test patch value
will have a chromaticity of D50.

So when you profile with that test chart, a spectrally flat
test patch will map the camera RGB to an XYZ with a chromaticity
of D50, so it will look neutral.

Graeme Gill.

Other related posts: