Digital BW, The Print

> 
> Dan Margulis devoted the last eleven pages (309 - 320) of Chapter 
15  
> of the 4th [Photoshop 7] edition of his "Professional Photoshop: 
The  
> Classic Guide to Color Correction" (Wiley, 2002) to an earlier  
> controversy about 8 bits per channel vs. 16. He didn't find 16 
bits  
> better in his attempts. But he did, on page 318, "... recommend 
that  
> if your scanner or digicam _can_ give you a 16 BPC file, take it.  
> Thereafter, I see no point in maintaining it in 16 BPC, but it  
> doesn't hurt either." I can't tell to what extent he was thinking 
in  
> the context of color prints or B&W.
> 
> Earlier in this section ("The Emperor's New Clothes"), Dan 
Margulis  
> wrote that he had "... no intentions of lavishing this much space 
on  
> the topic in subsequent editions." But I can't confirm that he  
> hasn't; the now five-year-old 4th Edition is the most recent one I  
> have, and a newer one might be a better reference for background 
on  
> trying out 16 BPC.
> --
> Sam
>
Most of the people here have probably heard about Dan's 16v8b 
challenge and know that till now no one has really proven 16b is 
superior. But this is mainly due to Dan's initial assumptions namely 
a sRGB working space for the file to be output on press, and he can 
not be wrong in this case. For those of you who follows this, this 
paper can be of interest : 
http://www.media.hut.fi/~as75192/tenttima/Ext_gamut.pdf#search
This tends to prove 8 bits suffice in this configuration. Recently 
Dan admitted for wide-gamut color printing that some images benefit a 
larger space, and the debate can rage again.
Now Dan has also always mentionned that correcting a greyscale file 
might call for 16bit for you do not get so many codes to describe 
shades and gamma 2.2 spaces might be best suited.
In the case of a scanner the 16b advantage is else : say your scanner 
is capable of real Dmax 4 (log 10,000 ; 10), then to encode that Dmax 
in bit you need puissance 4^2 hence 16b. So theoretically 8bit gives 
you Dmax 3 ; theoretically because 1 bit encodes whatever Dmax you 
want, but here we assume the bits are perfectly distributed, plus 
some get lost in noise. 
Yet both Tyler and Roy demonstrated that this is a pointless 
discussion when you have the print in hands : so you might benefit 
from a 16b correcting process, but nothing proves AOT the printing 
will benefit from a 16b file feeding. If I'm not wrong Krawitz turned 
Gutenprint in a 16b driver but not for the sole encoding of the 
values, but also to encode variable drop size, ink position, ect...at 
the end of the dithering not so many bits are devoted to the values 
themselves.

Olivier