Jim Kaye wrote:

>> Is this *with* or *without* boosting the impoverished blue channel
>> signal found in that environment?

>I hear what you're saying, Tom, but the thing I've never understood
>about using filters with digital like this is that the blue filter
>cuts down the transmission of other (i.e., red) wavelengths, it
>doesn't "boost" the transmission of the ones you want (blue). So
>unless you increase the actual exposure (slower shutter speed and/or
>wider aperture) when you use the filter, I don't see how you would end
>up with any more blue photons hitting the sensor than without the
>filter. And if there's enough light to increase the exposure, then why
>not just do that?

True, a color-correcting filter never by itself *increases* the photons
hitting the sensor over a given unit of time. It only cuts filters out
those of lower frequency hitting during that time.

Here's why you want to do that.

Let say you're capturing in incandescent light. Household fixtures are
nearly as cool in Kelvin as Nikon lets us enter without taking a manual
measurement. There's nearly no blue information there.

NB: You're going to have to look at this using UniWB so that the red and
blue coefficients are 1, letting you see true data capture.

With all blue channel data bunched up in the 1st quartile of the
graph, when you white-balance by multiplying the blue (usually also
red) components of the graph, two bad things occur.

The first bad thing is that your red component, which was not clipping
in the raw sensor data, runs the serious risk of doing so after scaling
by the white-balance coefficient, Previous to white balance, the red
component peaked about 60% into the graph. With white balance, it peaks
about 80% in, and old data to the right of 80% pushed way up and
eventually off the 100% edge. This results in hue shifts and detail
loss. Hold the exposure down to avoid that, and you get virtually no
blue data.

The second bad thing is worse. The blue component occupying the 1st
quartile of the graph represents *very* little data. What data is there
reaches a height of 6-8% of the graph's height until about 15%, and from
15% to maybe 30%, it's the thinnest flat line at the bottommost edge. As an
area, this is about 1% of graph's total normalized data representation. You
don't have much quality left in 1%; not much subtlety.

Scale by white balance, and you multiply that scant blue data by 3-4x,
sometimes more. This boosts the top of blue data; its amplitude increases.
Your little blue bumps in the 1st quartile push up toward the top, and
the thin line in the 2nd quartile gets expanded into a wrinkle. Your blue
now seems to cover covering maybe 15% of the available area. But remember,
your *real* data covered only 1%.

Imagine a deflated balloon with fine, well-rendered text printed on it.
Inflate it, and that print gets stretched into funny (=fun-house)
proportions. That's a bit like what you're doing. It grossly exaggerates
small differences, making things stand out that would, quite literally, be
unnoticeably lost in the noise--absent the distending amplification.

When you use a color compensating filter, you suppress the dominant
frequencies of the light, holding them back so they don't overexposure as
you gather the frequencies that *are* underrepresented. High altitude has
too much blue; daylight, too much green; incandescent, too much red.
Choose a color-correcting filter of a density that works for your camera
and light. I believe that Julia has suggested CCM20 may be enough on a D2H
for daylight work, but other numbers for other bodies.

http://www.aeimages.com/learn/color-correction.html
http://www.pochtar.com/NikonWhiteBalanceCoeffs.htm

So yes, Jim, you're right: this means a longer exposure. But by color-
correcting your light before it hits your camera, you'll get real data,
not false data created by amplification.

And you might not lose a full 2 stops. True, with a blue-admitting filter,
you'll pay more in exposure because blue contributes least to human
luminance perception, The precise density you'll need varies according to
your model's CFA dyes and shooting light. Maybe an 80A's 2 stops; maybe an
80C's one stop or an 82C's 2/3 stop will suffice to put you where digital
amplification won't cause too much noise.

Now think about what ISO boosting (read: pushing) does. You amplify the
data even more. Sometimes this amplification is an analogue signal
boost; sometimes it's simplistic digital multiplication of data (that's
what WB *always* is); and sometimes it's a combination of both, depending
on body and ISO setting. This exerts even further strain on the
extremely scant data that comes in. Add in people trying to dodge dark
areas. Put it all together and what happens?

You end up pushing/boosting yourself TEN FULL STOPS or more, and it
looks like completely curdled crud and mottled mush. Of course it does!
Yet people are surprised and astonished, shocked and dismayed--even, in
these parts it seems, enraged--by the noise resulting from such
ludicrously high pushes. They should not be. No sensor in our cameras
(today) gives great data pushed 10-12 stops. Expecting it to do so is
expecting pigs to fly--on a flat-earth world. Faerie-tale mentality.

Once we understand what's really happening, we compensate for what
nature doesn't give us on its own by using light-balancing filters at
the source. We stop living in a fantasy land that places unreasonable
demands on the underlying physics. With adult understanding comes
practical solution. Instead of behaving like some petulant toddler
endlessly complaining about the impossible, we find work-arounds that
are possible today without descending into faerie tale.

And we get far better images because of it.

--tom