Digital BW, The Print

This 8-bit vs 16-bit discussion keeps coming up but the assumptions made are
always over-simplified.   If you just "do the math" and say 8 bit means there are
only 256 grays and 16 bit gives 1000s of grays it's real easy to fall in the trap of
claiming that 16 bits are necessary.

Couple things to consider:

Any perception of gray -- whether it's our eyes or a high tech measuring device --
is over a multiple pixel area.  Take a smooth gradient and counting number of 
grays is almost entirely a factor of the specs on the measuring device.  
Use your eyes and you'll count something like 100 grays, use a densitometer 
that reads 0.00 to 1.64 and you'll get 164 grays, use and eye-one and it'll be 
a 1000 grays.  This is regardless if you use an 16 bit directly or convert it to
8 bit before sending to the driver.

Inkjet printing, especially, requires dithering over an area for perception of grays.
The printers take no where near 16 bits of data.  The most any of the Epson 
printers take currently is 2 bits for any ink.  In fact all the of K7 and K6 setups
use only 1 bit -- none of the setups use variable dot size.  The point being: what
difference does it make to go 16 bit to 1 or 2 bit directly or use the 8 bit value
as an intermediate?   As can be seen from Tyler's print scans, number of gray
inks has one of the biggest effects.   For a K7 ink system you can probably 
estimate about 4 bits or 16 possible grays at each position -- based on a combo
of about 2 inks at any point. But this is still way short of the 8 bits you feed the 
driver.  

I think once you get to the point where adjacent drops of ink are bleeding
into each other on the paper the result is no longer really digital or quantized. 
It's much closer to a continuous, analog image like a silver print made of chunks
of silver.

I'd venture a guess that most all "real" pictures have less than 8 bits of "real data".
I.e. by the time you finish editing the image the low bits of high bit data are
mostly noise.   This isn't to say that you can start with only 8 bits in a scan for
instance because which are the good bits is in different in the shadows and
highlights.  I do mostly 4x5 film scans and the smoothest gradients I ever see
still appear fairly noisy in 8-bit histograms.   Digital cameras are less noisy but
the bottom bit of 8 bit data is typically pretty noisy.

Unfortunately its a bit difficult to demonstate all this in any foolproof way. So I
guess the 16 bit discussion will persist and IMHO the main benefit of getting a full
16 bit throughout will be to take the issue off the table.

Roy

--- In DigitalBlackandWhiteThePrint@yahoogroups.com, "Steven Karafyllakis" 
<stevekphoto@...> wrote:
>
> David;
> 
> > Well linearized output of well optimized images, both at 8 bit, 
> have about as 
> > many levels as the eye can distinguish. 
> 
> This is excatly the assumption I'm questioning; OK, so if you laid 
> the steps out in a row, perhaps most people couldn't distinguish 
> from one to the next. If you laid out all the tonal steps in an 
> average 4x5 or 8x10 neg you certainly couldn't distinguish them, 
> they would appear continuous even though theoretically since the 
> image is made of dots (silver grain) it isn't really continuous. And 
> yet in terms of nuanced subtle gray transitions and tonal richness, 
> the difference between an inkjet printed in 256 steps and an 
> SG "continuous tone" print from 4x5 neg or bigger, is usually quite 
> obvious. It comes back to sheer information density, and a 256 shade 
> gray-scale can never quite reach it.
> 
> 
> High bit is wonderful to allow for 
> > adjustments, be it to capture, process, or output, but once things 
> are optimal, 
> > it is, for the most part, overkill. So I see high bit more as a 
> way to improve 
> > problem images or processing or printing than to further improve 
> top notch 
> > stuff.
> 
> I agree with your evaluation in this regard; but one thing I haven't 
> gotten clear on yet: does hi-bit equal more shades of gray, or 
> simply more control over the existing shades? And if the printer 
> driver converts back to 8-bit before printing, how does it help, 
> beyond, as you stated, allowing for corrections and more accurate 
> mapping of the existing shades?
> 
> What I'm suggesting, (wishfull thinking really) is killing the 
> sacred cow and setting up a system based on a more continuous gray-
> scale of 512 shades. A scanner function that would scan directly to 
> B&W in 512 shades. A monochrome digicam sensor that outputs 512 
> shades in raw format. Software that supports editing of the files 
> without downgrading to 256. And a printer driver that could make use 
> of that and deliver at least SOME of the extra info to the paper. 
> Remember, I'm looking for headroom, not neccessarily a quantum leap. 
> 
> 
> BTW, David, when IS PFP 2.0 going to be released? 
> 
> Steven Karafyllakis
> 
>  
> > C. David Tobie
> > Product Technology Manager
> > ColorVision Business Unit
> > Datacolor Inc.
> > CDTobie@
> > www.colorvision.com
> > 
> > 
> > 
> > [Non-text portions of this message have been removed]
> >
>