Re: Flatbed scanners vs Pro Photo CDs
Re: Flatbed scanners vs Pro Photo CDs
- Subject: Re: Flatbed scanners vs Pro Photo CDs
- From: Jan Steinman <email@hidden>
- Date: Tue, 5 Dec 2000 14:46:17 -0800
Okay, I said I was outa here, but Bruce just won't let me go... :-)
Converting an 8-bit image to 16-bit mode gives the data a few more
points to snap to, but it doesn't give you any more data than you
had to start with...
I fail to see the difference between this example, and taking a CCD
sensor that is only capable of resolving some 4,000 different light
levels, and digitizing it into 16,384 levels! Noise is noise! You can
"make up" data in hardware or software! Are you claiming that "making
up" the data in hardware is somehow better? THAT would be daft! :-)
With Bruce's introduction of $16,000 scanners to the discussion,
we're getting far from the original argument. I will accept that the
engineers who build a $16,000 scanner have had some systems
engineering training, and know where to put appropriate effort in
order to produce a useful system-wide response. I also know that if I
had designed a 120db IF amp to go after a 100db front-end, I'd have
been fired.
But if a 14 bit ADC costs little more than a 12 bit one, hey, why not
use it, even if it's input is not capable of resolving to 14 bits?
ESPECIALLY if marketing can put "14 BIT" very big on the packaging?
BTW, what prompted me to get into this in the first place was your
statement that "A "real" 14 bit scanner should achieve Dmax circa
3.6." It's a nonsensical statement.
Well, please let me qualify that a half-dozen different ways.
Substitute "real" with "well designed," or "engineered," or "not
driven my marketing specmanship," or "honest," or "credible," even
"useful." I apologize for using the ambiguous term "real." :-)
Hey, this isn't rocket science. You do the math. A 14-bit ADC
resolves 16,384 different levels, which is a potential
signal-to-noise ratio of 42db. Heck, I'll even spot you the least
significant bit and call it 39db. Inexpensive, consumer-grade CCDs
have a dynamic range of 33db or less. You're still digitizing 6db of
noise, which happens to be equivalent to two bits. A consumer grade
scanner is incapable of making use of more than about 12 bits --
period!
No matter how long you stare at the specs of an individual component,
it still won't tell you the system response.
One point we're probably confusing is "dynamic range" with
signal-to-noise ratio. The two are intimately related, in that you
CANNOT have a dynamic range that is greater than allowed by the
S/S+N, because noise is additive at every point of the response
curve. You may claim that you can resolve between level 8192 and
level 8193 of a 30-33db CCD from a 14-bit ADC, but in reality, that
level is uncertain to within the noise floor range, which for
consumer-grade CCDs, may be somewhat better than 0.05% of total range
-- the output is not repeatable and non-monotonic when sampled at a
resolution finer than that of the noise floor.
But hey, don't take my word for it! Go get that $399 scanner, scan
the same blank sheet of paper ten times at full resolution and 14 bit
output, go to exactly the same pixel in each image, and compare the
high-bit RGB values. They will have a variation of at least 0.025%,
which corresponds to a quantization level of 12 bits. (Unfortunately,
I just tried this in Photoshop 5.5, but even in high-bit mode, it
only tells you the 8-bit RGB values.)
I've been working with high-bit scans for about 10 years now,
So what? I worked on my first synthetic aperture digital sonar system
in 1979, designed my first digital sampled radio receiver in 1982,
and worked on the system software for CCD imagers for astronomy in
the late 80's. Has this descended into a "mine is bigger" contest?
When you can't argue facts, pound the table?
What is it about "you can't make a dollar's worth of bits out of
twenty-five cents of data" that is so difficult to understand?
Okay, I'm REALLY outa here now. :-)
--
: Jan Steinman <
mailto:email@hidden>
: Bytesmiths <
http://www.bytesmiths.com>