Digital BW, The Print

> Too much of a generalization.  You can get
> exceptional B&W from a scanner, which most
> all use RGB sensors.

The scanner isn't the original imaging device.  It doesn't matter what
colors it examines, as long as it can recognize light and dark, because the
material being scanned is colorless.  Even a single scan in purple would do
just fine.

> Also, scanning backs use RGB and give full
> color data, and they convert very very well
> to grayscale.

Not as well as if they just scanned in black and white to begin with.

> Well, if you're talking cameras, Kodak has,
> and if you're talking scanners, the Leaf was.

Kodak's camera is the rare (unique?) exception to the rule.

> Have you actually done any experiments, or do
> you know of any on the web?

I've done both real and thought experiments.

It's easy enough to see the theory.  Start with any RGB image, and try to
reconstruct the original spectral distribution of the light that produced
it.  You'll find that this is impossible, as more than one distribution can
produce any give RGB signal.  Because of this, no RGB signal can be
converted to grayscale in a way that will emulate any arbitrary original B&W
image, because original B&W capture is a function of a continuous, often
irregular curve of spectral sensitivity, whereas an RGB image is just three
numbers.  A great deal of information is thrown away in RGB capture.  The
same is true in B&W capture, of course, but as long as the B&W is the
_original_ capture, it doesn't matter--you won't be trying to convert it to
grayscale if it already is grayscale.  (Note, however, that converting one
B&W image to another is even less plausible than converting an RGB image to
an emulation of an arbitrary B&W image.)