On Monday, September 22, 2003, at 06:09 PM, Brad Davis wrote:
> On 22/9/03 16:44, "Roy Harrington" <roy@harrington.com> wrote:
>
>>
>> Brad,
>>
>> You are probably right that -- "This conversation is only about bit
>> depth."
>>
>> But I think that is one of the main shortcomings of the discussion.
>> Resolution
>> (DPI or more accurately PPI) and bit depth are certainly very
>> different
>> and
>> clear properties of an image file. However the criteria for
>> comparison
>> is always
>> a print that the human eye looks at to evaluate. At this level PPI
>> and
>> bit depth
>> are no longer independent and well defined properties of the image.
>>
>> Fewer levels of gray (i.e. less bit depth) is easily compensated for
>> by
>> a higher
>> resolution --- that's the basic principle of halftoning or of
>> stochastic dithering in
>> printing.
>>
>> Roy
>>
> Roy,
>
> First, I agree, I didn't think that the post I was responding to was
> about
> PPI (Thanks, that clearly is more apt, although Polaroid rated my
> scanner in
> DPI - but what do they know, they went bankrupt!). Adding the issues
> you
> introduce might make the case for 16 bit a little more certain - I
> know that
> I started out with a strong bias toward 8 bits for a number of
> reasons, not
> the least being the human perceptual system, but there are subtle
> things
> that keep coming up which make me think I may have been wrong - the
> interaction between PPI and bit depth being one - I was also able to
> think
> of an example of how a color image might show banding (Austin's
> arguments
> not withstanding), even though Henk refused to support his opinions
> with an
> example.
>
> I still can't figure how a scanner can manage to be noise free
> (electronically) at 1/65,000th of its voltage range (that would be
> necessary
> for a full 16 bits - but I can imagine a somewhat smaller bit size as
Brad,
I think it's totally a fallacy that a spec of 16 bits means the scanner
is that noise
free throughout its range. A significant portion of the noise will be
proportional
to the size of the signal. So while its likely that you can
distinguish values
34, 35, 36, 37, when you get to the large numbers like 10000 it may be
that
the next distinguishable value will be 10100. (That would be noise <
1%).
If you figure on 16 bits and 1% noise you get somewhere near a maximum
of 600 distinguishable values. Take into account the fact that you
don't
use much of the total possible range and you can get down to 256 quite
easily.
So in changing from 16 to 8 bits its mainly a matter of whether or not
you did
well in mapping to an "interesting" set of 256 values i.e.
distinguishable in
the scan and desirable in the print. Secondly with enough resolution
you
get some wiggle room because Photoshop dithers between values on
conversion to 8 bit.
Roy
> working (12- 14?). I also wonder how much dithering (effectively
> increasing
> the apparent bit depth by increasing the number of levels between
> adjacent
> pixels is done by say Epson or HP - or whoever.
>
> Also, in all of this, I haven't seen any reference to the limitation of
> paper to present more levels on the one hand, and the subtle effects of
> increased levels on perception. Simple psychophysics isn't sufficient
> to
> explain all of our perceptual capacity - photography paper (silver)
> could be
> much cheaper is we psychologists had it all locked up.
>
> brad
-
Roy Harrington
roy@harrington.com
Black & White Photo Gallery
----------------------------------------------------------------------------------------
Unsubscribe by mail to listserver@halftone.co.uk, with 'unsubscribe
filmscanners'
or 'unsubscribe filmscanners_digest' (as appropriate) in the message title or
body
