Jawed wrote:
>An 8-bit A/D really would struggle.
I agree but it was as I mentioned, an artificial example. Maybe I should
have worked with what I actually have, which is a scanner with a 12 bit
A/D that the firmware drops out the 2 LSB from to return 10 bits per channel.
It doesn't matter how many bits are on the A/D - it could be 32 for all
it matters. What matters is how large a range of numbers is produced.
With a neg film, the range will be much more narrow than a slide film.
If it equates (by your estimates of 6 to 7 bits) to a range of only 5 to
6 bits of variation, the output will posterise easily.
> 7.x bits of scanned data are gonna be stretched a bit
> thin by the time the image has been "decoded". Even
> more so when you start doing Levels/contrast
> adjustments or curves.
Exactly. So you need as many bits per channel as possible on the input
side to get the best quality at the other end.
[CCD linearity]
> "integration time". Whether this effect is ever significant
> is something I don't know. Really needs someone who's
> implemented a scanner at this level of detail to comment.
In the past it's been stated on this list that the CCDs used are linear
for all intents and purposes.
>> sensitivity compressed into a much more narrow range. If you don't
>> have enough bits, you won't get the subtle tones whichare in the film.
> Agreed. Though this does depend on how far from the desired result the
> scan is.
Sure. The less you have to manipulate it, the better - yet isn't this contrary
to experience? Most people seem to be saying that it's easier to get good
scans from colour negs than slides, yet they require a LOT more manipulation
of the data than scanning a colour slide.
> When you scan a true b+w film, it seems the scanner loses a
> very significant amount of dynamic range, a bit or more (I
> don't know what the mechanism for this is - is it solely
> because scanning is not done across the R, G and B
> channels - but only on, say, G?).
I don't know what Nikonscan does, exactly - it seems to take the output
of only one channel, and only as 8 bit data. That immediately cuts the
dynamic range. Vuescan combines the colour channels into a high bit file.
But AFAICS it's the density of B&W film which is the problem fot a Nikon,
aside from nasty interactions between the LED light and the silver grains.
>> No, they don't. As Ed explained, they change the exposure time not the
>> brightness of the LEDs.
>I believe you're misquoting Ed.
No. Ed posted very emphatically that he had traced the commands used by
the Nikon scanners, and they changed the integration time, nothing else.
He also mentioned that changing the "analog gain" affected the speed of
the scan. If it was LED intensity, the scan time would be unaffected.
> The manual says the brightness changes, too
The manual is misleading. Either the tech writer was trying to explain
things in photographic terms of exposure compensation or they were misinformed.
> Unless you mean that there is no connection between
> Nikon Scan's auto-exposure algorithm and LED brightness.
There isn't. The LED brightness is fixed and the exposure is adjusted by
varying the integration time.
Rob
Rob Geraghty harper@wordweb.com
