on 9/2/02 12:20 AM, Austin Franklin at darkroom@ix.netcom.com wrote:
>
> Roy,
>
>>>
>>> I see how size can have a merit (which is a relative ratio),
>> and range, as
>>> they apply to dynamic range. Size in the fact that the largest
>> signal is N
>>> times larger than the smallest...and range in that you can say
>> "all integer
>>> values from 1:1 to N:1". BUT...realize that "all integer
>> values from 1:1 to
>>> N:1" really denotes a resolution over a particular "range"
>> too...that you
>>> have N discrete values.
>>
>> Yes, but I never said "integer". In the real-world i.e. analog, there no
>> reason why any real number couldn't be used. What's wrong with going from
>> 1:1 to 1.01:1 to 1.02:1 ...
>
> Because noise is 1, and you can only measure in increments of noise. In a
> system that has noise of, say, 1V, you certainly can't measure 1.01V, now
> can you?
You probably haven't seen the post yet. But, I've actually demonstrated
that you can in that post about "noise" and the TIFF file that you can
easily print out and measure with your densitometer.
>
>> Here's why I have a problem with the "concept of resolution":
>>
>> Let me go through a simple example of a (semi-idealized) scanner.
>>
>> Here's the basic specs of the scanner:
>> Density Range: 0D to 3.6D
>> Bit Depth: 12 bits
>> Number of levels: 4096
>>
>> A couple of simple observations:
>> The density range is also 12 photographic stops -- each stop is .3 of
>> density so 12*.3 = 3.6
>> You can chop up the density range into 12 one-stop ranges i.e.:
>> 0 to .3, .3 to .6, .6 to .9 ... etc to 3.3 to 3.6
>>
>> Photographically and human perception wise each of these one-stop
>> ranges are equivalent in size.
>>
>> So now let's chop the density range into the 4096 levels. The
>> density range 3.6 divided by 4096 gives a little less than 0.001D
>> per level. Approximately, 300 levels for each of the 12 one-stop
>> range. Sounds like a great concept of resolution, doesn't it?
>> We get a new level every 0.001D change in density -- it sure
>> looks like a resolution of 0.001D.
>
> But that's not how scanners work. They know NOTHING ABOUT density values at
> all! They only know photons, and how many photons the CCD sees. They see
> relative values output PURELY AS A VOLTAGE (or possibly current), and that
> voltage has a range, and has noise. You can only measure as accurately as
> noise, and as such, noise defines the resolution of that system.
>
> <snip>
I really wish you'd read what I write and not snip out of context.
Immediately after the above paragraph I wrote:
Fine, but the trouble is: scanners don't work anywhere even remotely
close to that scenario. This is what scanners actually output
as the levels:
The first one-stop range contains 1/2 of all the levels, the next
on contain 1/2 of the remaining ones, etc. until the last one.
>
>> Austin, don't take my word or the web's word for it. Try it yourself.
>
> Roy, I've designed film scanners, and have been designing digital imaging
> systems for over 20 years. I KNOW how they work.
You may know scanners inside and out. But with scanners, resolution
and levels are all based on QUANTIZATION noise or more accurately
QUANTIZATION ERROR. You may pick this quantization based on the real
random noise of the system, but all the properties about resolution and
number of levels are based on the properties of quantization.
The true random noise of the input signal has VERY DIFFERENT properties.
Quantization noise is a hard limit on resolution, but RANDOM noise
presents no such boundary on resolution. With multiple samples
that are averaged you can increase your resolution accuracy.
---------
This is why they have some high end scanners and software that take
multiple scan passes. The scanners are designed so that quantization noise
is much less than the input signal random noise. With a single pass
the random noise is the limiting factor and the bottom couple of
bits in the output are just noise -- no information. With multiple
passes you average several samples getting a more accurate measurement.
You are increasing your resolution beyond the random noise level; when
you get down to the quantization level its a hard limit on resolution.
That's the best you can do.
>
> Austin
>
Roy
Roy Harrington
roy@harrington.com
Black & White Photography Gallery
----------------------------------------------------------------------------------------
Unsubscribe by mail to listserver@halftone.co.uk, with 'unsubscribe
filmscanners'
or 'unsubscribe filmscanners_digest' (as appropriate) in the message title or
body
