Don said:
Well, compensating for linear gamma is still editing in the broader
sense and that was implied.
Why? Until relatively recently (10-15yrs) all scanners produced a
digital output directly in gamma compensated space. I would argue that
the "editing" you refer to above is merely a preprocessing requirement
of the technology used in modern scanners, similar to dark current
compensation, necessary to generate the information in the standard,
displayable, form.
Isn't there is still a problem of editing the image afterwards?
Editing in 8-bit could still cause banding - at the very least in some
extreme images - so, in that respect, the 16-bit elbow room would
still be indispensable.
That is the theory - Dan Margulis issued a challenge some years ago for
anyone to demonstrate it. To date, nobody has. Perhaps you would like
to try to be first? Simply create an image working from a hi-bit depth
original, gamma compensated film scan by processing in 16bit per channel
mode which cannot be matched by processing in only 8-bit per channel
mode throughout.
Note that the 8-bit process steps do not have to be the same as those of
the 16-bit process, merely produce the same end point - so steps such as
applying ridiculous compression in 16-bits only to re-expand the data
into a displayable form, would simply be bypassed on the 8-bit process.
Over to you, Batman!
I wonder why consumer scanners don't do that? As you mentioned once
before, it's pretty elementary to implement. Yes, hardcoding gamma
appears less flexible at first blush but since gamma of monitors and
perception coincide around 2.2 it would seem like a natural choice.
Not to mention it can always be made selectable and would certainly
make life easier for most people.
I suppose it must be those Mac users again with their urge to be
different with their 1.8 gamma. Form over substance, indeed! ;o)
Nothing to do with Macs, Don - the difference between 1.8 and 2.2, in
terms of the effect on posterisation, is negligible. The reason this
isn't done in CCD based scanners is primarily because of something I
mentioned above - dark current compensation. The dark current present
on each CCD element is significant and unique to each element and would
produce the effect of some columns of the scan having visibly more
effective levels than others if the signal was digitised after an
analogue gamma compensation step and the dark current removed in gamma
compensated space. As I explained to you during your own process
development, removing dark current in gamma compensated space is not a
simple subtraction, but some fairly complex arithmetic functions which,
in this sequence would require very high precision to prevent visible
errors in the result just as a consequence of analogue noise on the CCD
output. It is much simpler and faster to remove the dark current from
the CCD output in linear space and then convert the resulting signal to
gamma compensated space afterwards. If there was a simple, cheap and
effective way to suppress the dark current in the CCD itself and flat
field them so that they produced a uniform output then it certainly
would be worth applying the majority of the gamma compensation prior to
the digital conversion step. Since there isn't, the cheapest and
simplest approach is just to throw more bits at the problem, digitise
linearly, dark current and flat field in linear space and convert to
gamma compensated space, in that order. However, as you know yourself,
you need more than 17-bits of ADC range in linear space just to achieve
the same range as an 8-bit gamma compensated range.
You can imagine the conversation in the product pre-release conference:
Engineers: The new scanner has a 20-bit ADC, the signal is gamma encoded
and output at 8-bits, this uses the full 17.5-bit linear range with some
margin for noise and dark current suppression, so that the scanner
delivers true drum scanner performance in 8 bits per channel.
Marketing: Why don't we give the customers all 20-bits?
Engineers: They don't need them, they can't see them and most of the
range is used anyway to produce the high quality output in perceptual
space.
Marketing: But Nikon, Canon and Minolta sell their products on the
number of bits they output - we won't compete if we only give the
customers 8 of them, when they have twice as many.
Engineers: But we are using our bits smarter, so we just need some
smarter marketing to... err, right, we see your, um, problem. OK, we'll
output an extra 8-bits and fill them with random numbers so they won't
compress losslessly, nobody will know the difference till their disk
fills up. We are scheduled to launch our new disks a month after the
scanner...
Marketing: I love it when a plan comes together. ;-)
