Digital BW, The Print

Austin writes:

> To do the conversion from RGB to grayscale, you
> are not reconstructing a curve with the three data
> points ...

Yes, you are.  The response of a B&W sensor to the original scene is a
function of the complete curve present in the original scene.  To simulate
that response, you must have the original curve.  Therefore, to get that
response from an RGB image, you must have a way to accurately reconstruct
that curve from your RGB values.  Every possible original curve, in other
words, must have a unique RGB representation.  If there are any
duplications, then it will not be possible to construct the B&W image from
RGB data alone.  And unfortunately, in practice, there are many
duplications.  Each RGB value can represent an infinite number of different
original curves.  And since these original curves do not all produce
identical results on sensors with different spectral sensitivities from the
sensor that produced your RGB data, you cannot reproduce the B&W results
from your RGB information.

> Frequency is one vector and
> intensity is the other.  That's it.

But the original scene didn't contain just one frequency.  It contained an
entire spectrum of frequencies, for _each point_ in the image.  And you've
lost those in your conversion of the original light in the scene to a
triplet of RGB values.

> You do this for EVERY different RGB value that
> is valid for the film type you are "simulating",
> and you have reconstructed the grayscale response
> curve for that film.

You can't.  See above.