Digital BW, The Print

Peter writes:

> We see yellow because the photopigments for
> red and green in the retinal cones both happen
> to overlap their sensitivity in the range
> of wavelengths we call 'yellow', between
> about 540 and 560 nm.

Correct.

> Clearly there is information lost in your
> understanding of the physiology of color perception.

No, my understand is correct.

> If a blend of red and green produces the perception
> of yellow it IS yellow.

To human eyes, yes ... but to other types of image capture devices, no.

I suppose I should return to my example of the yellow-sensitive film, the
one that everyone persists in ignoring.

Truly yellow light, from a physics standpoint, falls within a narrow band of
wavelengths.  It is distinct from the blend of red and green light that
human beings perceive as yellow.  However, human beings cannot tell the
difference between this "real" yellow and the "perceptual" yellow made by
mixing red and green.  Similarly, all color display and printing systems for
human beings are designed around this defect of human vision, and so they
cannot make this distinction, either.

However, it's certainly possible to develop an imaging device that is
sensitive only to spectral yellow, and does not respond to red or green.
Such a device, when presented with a scene that contains both
spectral-yellow-lit areas and areas lit in red and green, will see only the
spectral yellow areas as being lit.

Now, the problem arises when an RGB-based capture system records this scene.
The system will see all of the scene as being equally lit, because it cannot
distinguish between spectral yellow and a blend of red and green.  The RGB
scene thus recorded will therefore show all areas of the scene lit equally.

Given this, it should be obvious why the RGB scene cannot be transformed
into a correct representation of the scene as it is recorded by the
yellow-sensitive device described above.  The distinction between spectral
yellow and blended red and green, which would have been recorded by the
yellow-sensitive device, disappears when the scene is recorded in RGB.
Thereafter, no amount of manipulation can change this.

This illustration is a very simple example, but the principle applies
equally in much more complex situations.  And this is why one RGB
representation cannot generally be transformed into another, nor can it be
transformed into B&W representations with complete accuracy.  Too much
information is gone.

> If you take a photo of a scene with an RGB monitor
> in it with a yellow screen you know that the yellow
> is created from red and green phosphors.

Yes.  But if there is a yellow box next to the monitor, you do not know
whether the box is truly yellow (that is, reflecting light at the frequency
of spectral yellow) or just appears to be that way because it reflects red
and green equally.

> Are you saying that black and white film will
> record this differently from an identical yellow
> created by say mixing cadmium yellow paint?

YES!!  (If cadmium yellow is a spectral yellow, and I think it is.)

B&W film--as well as color film and CCDs--records light differently based on
its frequency, not based on how it might be represented in RGB.  If the
sensitivity peaks at spectral yellow frequencies but is lower and unequal at
red and green frequencies, the B&W film will render different "types" of
yellow with different luminosities.

This is what gives different films and different imaging devices their
distinctive palettes.  And this is also why no accurate conversion between
palettes is possible using simple RGB information alone.

> The only way what you're saying could occur is
> if the sensing elements (whether film dyes, photopigments,
> or CCD filters) were NONoverlapping.

No, it happens even for overlapping sensitivities, as I have explained for
the umpteenth time above.

> Imagine an RGB sensor that ONLY saw 640 nm, 500 nm
> and 440 nm.   If you had something in your scene
> at 560 nm you would see yellow, amd TMax film would
> record a shade of gray, but the RGB sensor would see
> black.

Correct.  But this principle also holds when the RGB sensitivity is a group
of curves, instead of monofrequency sensitivity as in your example.

> What you're probably thinking of is metamerism.

No, what I'm thinking of is what I've described here in detail, over, and
over, and over.  Isn't there anyone who understands this?

> Using the above numbers, if I have LEDs of 440,
> 500, and 650 nm illuminating a color patch that
> only reflected at 560 nm it would look black.  It
> would also photograph black with all film and with all digicams.

Yes.  But I'm not talking about illumination, I'm talking about capture.