Digital BW, The Print

Sorry for the long post, but I thought I might add what I've discovered 
with my images. After reading through this thread about high bit image 
editing, I thought it was worth noting that one important distinction 
needs to be made. I've spent some time experimenting with 8 and 16 bit 
BW scans and trying to determine if there is a perceptible difference 
in the output image.  Also have had numerous discussions with Antonis 
on this, who's knowledge of PhotoShop is truly phenomenal, and we've 
recently done some testing with hi-res piezo's output on the 7000. Here 
is what my current conclusions are, based on what I've been able to 
see:

There is no doubt that when you drop from 65,000 levels of gray (16 
bit) to 256 (8 bit), there are pixels that shift in tonality, since 
they are moved to the nearest 256 level gray value. But whether or not 
you will see this, either as a slight compromising of the tonal range 
or as flat-out posterization has to do with the resolution of your 
scan, if it is a B&W film neg we are talking about.  Computer generated 
images are a different matter.  The key seems to me to be the 
distribution of pixels and their relative values to each other.  If, in 
16 bit, you have a large region of pixels that are the same value next 
to another large region of pixels that is one value lighter, you 
probably won't see a difference in a print, as they are only 1/65,000th 
of the tonal spectrum apart.  But if you convert to 8 bit, either PS 
will assign them to the same gray value OR to adjacent values depending 
on the choices it makes, and the two regions then become only 1/256th 
apart, and the border between these areas will be visible in print, ie. 
posterization.

However, and this is the important part, if your scan is high enough in 
resolution, at the pixel level you will see very few regions where 
there are many pixels very close together in value.  In the scan, each 
silver grain in the film is made up of numerous pixels, and these will 
vary substantially in value depending on whether they are located in 
the middle of the grain, the edge, the clear film base between, etc. 
You will see this if you look closely down at pixel level, the image 
actually looks "noisy."  But the perception of tone these pixels render 
is effectively an averaging effect, since we are seeing them from very 
"far away" in a sense, since our prints are not such severe 
enlargements.  So if you take a 16 bit scan and drop to 8 bit, those 
pixels will each shift marginally lighter or darker, but since the 
viewer's impression of tonality is made up of their "average," one 
usually cannot perceive this. And presumably the number of pixels that 
shift lighter and the number that shift darker will be relatively equal 
as this is a random process. Posterization will only occur if you have 
a predominance of one color pixel in a region that is visible at the 
viewer's level of resolution.  This can, BTW, develop in smooth tone 
areas of a hi-res image if you res down, or print _very_ small, as this 
may force pixels that were once only slightly different in value to 
become the same, again producing posterization.

I print a lot of skies, and in low-res scans I have trouble with 
posterizing if I use severe levels or curve tweaks in 8 bit. Or if I 
blur to smooth out grain. But I can get away with the contrast tweaks 
just fine in hi-res scans.  There is no visible difference in the 
output of an image that is 16 bit going to the printer and an 8 bit 
one.  Even though there is combing in the histo's of some of the layers 
in my PS files I can see no difference in print, which is what matters 
to me.

Hopefully this is coherent enough, it's a bit hard to explain using 
just words. Also hope it sheds some light...

Phil
http://philbard.com