[argyllcms] Re: White Point

  • From: Ben Goren <ben@xxxxxxxxxxxxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Sun, 8 Nov 2015 20:06:09 -0700

On Nov 8, 2015, at 4:01 PM, Hening Bettermann <hein@xxxxxxxxxxxxx> wrote:

I have been reading the Luminous Landscape site for many years, and it has
given me the feeling that much of the theory is unnecessary complicated,
because it is only necessary to de-construct what Adobe puts out to be

Yes, exactly. And it's why I've never spent much time at LuLa...there're some
great artists doing great work there, but in spite of their methods rather than
because of them. Or: study them for their artistry, but ignore their

Unfortunately, it is exactly math & physics which are foreign languages to me
:-( - but I'll try...

You're in luck. The physics and math involved are very, very simple. Ignoring
color, it's literally nothing more than counting photons in one way or another,
though often rather indirectly. Each stop represents a doubling or an halving
of the number of photons. From 1/60s to 1/30s is obvious. That exposures go
from f/1 to f/1.4 to f/2 to f/2.8 to f/4, with each being a stop and therefore
a doubling, is less obvious, but makes sense once you start calculating the
areas of circles and the like -- and not important in practice. ISO counting is
more obvious but the mechanism behind it is more opaque...but again doesn't
matter save that higher ISOs generally mean more noise (and thus less dynamic

When you bring color into it...you start with the full spectrum, and the light
can be any wavelength. Each pixel has one of a red, green, or blue filter over
it. Each filter will let through at least some photons of any wavelength, but
the red filter lets through most of the red photons and few of the blue ones,
and vice-versa. Your eyes do much the same thing. If the proportions of how
much of each wavelength made it through the camera's color filters and your
eye's respective color cells, there'd be no need for color management. But
they're not, and all color management comes down to, basically, mapping this
camera's spectral sensitivity to the human eye's...for example, most cameras
are much more sensitive to green than you are, so your exposure will often max
out the green channel and, in post-processing you digitally boost the ISO on
the red and blue channels to make up for it; that's, conceptually, the first
thing that you're doing with white balance. Next, you might or might not be
trying to make the scene appear as if it had been photographed under a
different illuminant other than what it really was, and that's the rest of
white balance.

Where a color profile comes in...it's not just the overall relative strengths
of the three channels where a camera differs; it's also much more fine-grained.
The peak frequencies are different for eyes and cameras. The shapes of the
curves are different. But the math is such that you can create a mapping
between the relative intensities of the three channels that the camera sees and
the relative intensities of what an human would see...and that mapping is the
color profile. For most cameras, you can create a very accurate mapping for the
kinds of spectra you get from reflective surfaces...but, the more saturated the
colors get, the more apparent the divergence between the camera's color
response and an human's becomes, and the camera is more likely to record a
given pair of spectra the same that an human would see differently and
vice-versa. That is, humans are color-blind for some colors that cameras can
distinguish and cameras are color-blind for other colors that humans can
distinguish. But that's rarely a problem in practice, since few if any printers
or displays are capable of reproducing those colors anyway.

Most of the rest of color science comes down to filling in all the details I
obviously just glossed over.

I am a bit, no a lot surprised about what you write about shooting the
ColorChecker in the field. I think that it's built in logic amounts to making
every image look like it was shot under standard light, i.e. eliminating the
color of the light as I saw it. But that is another discussion.

Ah...that's the basic intent of the software bundled with the Passport, as well
as lots of other profiling workflows. It's certainly what you want to do for
art reproduction, and much of the history and current practice of the field is
devoted to such.

But the real reason to photograph the ColorChecker is because it's a known
reference. It's just reporting the weather, so to speak, and it's up to you to
decide what to do about it.

There will be practical and artistic considerations...the white point of the
light you're shooting in will be different from the white point of your monitor
will be different from the white point of the lights in the gallery where the
print is shown. You can map the white point from the one to the other, you can
keep it constant, you can blend the two, and so on...if you know what you're
doing. For that matter, if you really know what you're doing, with the right
equipment you can take a picture under, say, noontime light, and make the
colors look as if the scene were lit by a campfire. Or vice-versa. (With
certain limitations, mostly related to camera noise and spectrographic

For now, it looks like I have to abandon Iridient in exchange for an easy
shortcut to a good camera profile...

Iridient is a good product, but it's a bit opaque and limiting when it comes to
getting directly at the raw data. Raw Photo Processor doesn't have anywhere
near as slick a user interface...but it gets the hell out of your way and
produces superlative results. RPP might not be the choice for those looking for
a quick-and-easy solution, and it probably doesn't do as well when it comes to
salvaging the effects of bad technique...in that sense, compare it to a race
car as opposed to an SUV. Both are good for what they're designed for, but not
so good at what they're not designed for.

And nothing says you have to use the one exclusively. I'm sure there're
champion race car drivers who take their kids to soccer practice in SUVs....


Attachment: signature.asc
Description: Message signed with OpenPGP using GPGMail

Other related posts: