Digital BW, The Print

Jon,

>  > This is all quite interesting, as MANY people convert RGB images to
>  > grayscale, and they look quite good.  If this was so inaccurate and
>  > horrible, then how DOES that work so well?
>
> Ah, kind sir, I never asserted that one cannot get great results from the
> process.  :)  In fact, I've printed a large number of quite pleasing b&w
> prints from color originals.  However, one of your original statements,
>
>  >I have all the information I need, the frequency and the intensity.
>  >Both the color and B&W films response is deterministic to the
>  >frequency and intensity...so I believe I have the information
>  >necessary to map one to the other.
>
> ..is a mathematical and practical fallacy.

Apparently not.  For two reasons.  One, the spectral information is
irrelevant, and two, the intensity information IS contained in the RGB data.

> Simple example:
>
> Let's say that you're using Kodak's sensor listed here:
> http://members.aol.com/modernimaging/mi/Kodak_DCS-620x_Technology.htm

Well, that's a Bayer pattern sensor, and isn't what is used in film
scanners...but fine, that works for your example.

> you have a nice blue
> object.. let's say it comes out to one very dominant at 470 nm.
> Let's say
> that it's about as bright as the whole of the scene.  (We do the latter
> just so we can skip over talking about development times, over- or
> under-exposure, etc.)  Let's say that right next to it you have another
> blue object at 430 nm, same brightness.
>
> On the CCD, these two blues excite the sensor the same amount.  They will
> both be "weighted" the same amount.  You'll get some output value B, with
> no output values R or G (their response curve is 0 at that point.. the
> argument holds even if they aren't).  Compared to the overall brightness
> of the scene, you'll get some number for B, say.. oh.. 100, for
> each of them.
>
> On the Tri-X, notice that it's spectral response curve is falling through
> the blue region.

Er, insignificantly...

> It will weight the 470 nm color less than the 430 nm
> color.  The first cel will get a brightness value (compared to
> the overall
> brightness of the scene) of, oh, say, 100, and the second will get a
> brightness value of, oh, say, 90.

Your numbers are fictitious.  How do you know it's not 1026 vs 1032, which,
as far as your eyes can tell, show NO difference?

> So does RGB(0,0,100) map to B&W(90) or B&W(100)?  Ah, there's the rub...

Only if you show that it's significant in reality, which I do not believe
you can.

> it maps to both, but you no longer have enough information about
> the input
> to be able to determine which.  Again: the mapping is not unique.

And it does not have to be!

> So, when you convert your RGB image to black and white to try to emulate
> Tri-X, are you going to map that triplet to 90 or 100?  Or any of the
> other numerous values it might be?
>
> Let me state again:  it is a simple mathematical and real-world-proven
> fact that one cannot convert the output of a binning process into the
> output of another non-identical binning process in all but the most
> trivial of cases (that is, trivial response functions, such as impulse
> functions).  I see this most every week at work.

And, as I've said, I believe that's not relevant.

> Again: this is *not* to say one can't get great b&w images by
> desaturating
> color images.  See my disclaimer above.  I love my trusty G2, my NPH, you
> name it.  However, one cannot claim to be able to reproduce sensor X's
> output by shaping the output curves of non-identical sensor Y, whether
> that is comparing film to film or CCD to CCD or CCD to film.  That's just
> bad science.

I believe you are mis-applying "science" here.  Let's just leave it to good
engineering ;-)  I've designed similar systems that work just fine, where I
reconstruct, via characterization, the original response.  I'm not saying
that works for every instance, but for THIS instance, I believe the correct
information is there in the RGB data to do so.

The thing that you're missing is significance.  Yes, there are differences,
but is this difference discernable in the final output?  I say no, and you,
or anyone else, has yet to prove that not true.

I'll go back to pointing out that development and exposure are CRITICAL to
what you get for a "look" out of Tri-X.  Given the WIDE variance that Tri-X
can provide, I believe that shows that I in fact have a wide variance to
which I can "simulate" to.

Austin