Bart van der Wolf said:
I should have elaborated a bit more on what I intended to say, and
should have rephrased to
"It is not optimally possible with a single white light exposure". I
had hoped that the snipped part made it clear that there are other
ways, but with their specific caveats.
IMHO a better way of tackling it is with electronic shutters
(
http://microscopy.fsu.edu/primer/java/photomicrography/ccd/shutter/ or
http://www.kodak.com/global/plugins/acrobat/en/digital/ccd/applicationNo
tes/ShutterOperations.pdf) per scan line, although we cannot be sure if
they are actually used. Especially since we are talking about a
consumer level scanner, I wouldn't be surprised if shortcuts are used
instead.
Those references seem to refer to 2-Dimensional CCDs, not linear
devices. A mechanical shutter is required for many 2-D CCD because they
use the same CCD structure to transfer the charge packets to the output
as they do to capture the image in the first place. Consequently, the
CCD is still responsive to the image as the charge moves through the
device. Whilst this can be beneficial in certain cases (time delay and
integrate, for example, in a scanning system) it generally results in
significant image smearing.
In a linear CCD that doesn't happen. The charge is transferred to a
parallel CCD on completion of the exposure. This second device is
'blind', and the charge can then be read out without any smearing of the
image. Consequently there is NO need for any "mechanical shutter" with
a linear CCD - because each line is actually two lines, an image line
and a transfer line.
There are basically four ways (I can think of) of controlling the ADC
output levels.
1. Different (illumination levels or) integration times with a single
scan line.
As used in (all?) film scanners.
2. Different pre-ADC offsets and/or analog gain per R/G/B color.
3. "Electronic shutter" control per R/G/B scan line.
Defaults to version 1 for the linear CCDs used in scanners.
4. Post ADC processing.
No.1 is very feasible with single scanline or full image matrix sensors
and R/G/B colored light, and it will 'optimize' speed because only the
more attenuated channel colors need longer exposure.
It is also eminently feasible with a single white light. Each of the
RGB lines have independent integration controls. When the CCD line is
held in reset mode, all of the photocurrent generated is effectively
dumped to the substrate, not into the wells. Turn reset off on any line
and the device starts to generate photocurrent signal. The red, green
and blue exposure can be controlled separately even with a single white
light source - it does not need separate R/G/B lights - the CCD lines
are individually filtered to the three colours.
Only in Nikon scanners does the exposure for each colour takes place
sequentially on the same CCD line(s - each responding to the same colour
of illumination at any instant). The exposure time can be different
for each colour because each colour is sensed sequentially.
Conventional scanners use a white source and filter at the CCD - the
exposure of each filtered line is controlled independently.
No.2 is very likely to introduce color shifts at certain film
densities, due to non-linearities through-out the full ADC range
(especially near the saturation level). In any case it would require
stable operating temperature control and frequent calibration, but
still increase noise for the "weaker" channels.
And it wouldn't work any better in practice that Version 4 because most
CCDs don't have a noise floor that is significantly better than the ADC
performance of most modern scanners. In the case of the Minolta, I
seriously doubt that the CCD has a noise floor that is better than the
16-bit ADC - so you won't get any benefit from analogue processing prior
to the ADC.
No.3 may or may not be available for the sensor used (it seems more
common for certain types of cameras and implemented as global
electronic shutter for the entire frame), we don't know whether it is
for this scanner.
Every tri-linear colour CCD I have ever seen has this capability.
For example, a fairly old but typical device at:
http://www.kodak.com/global/plugins/acrobat/en/digital/ccd/products/linea
r/KLI-2113LongSpec.pdf
See Section 1.4 - there are separate LogR, LogG, LogB controls to effect
independent exposure control on each colour.
The complete Kodak range is at:
http://www.kodak.com/global/en/digital/ccd/products/linear/linearMain.jht
ml
The complete Sony range is at:
http://products.sel.sony.com/semi/ccd.html#CCD Linear Sensors
Individual control of the Sony CCDs is via the ROG control.
All tri-linear CCDs have similar independent exposure control for each
line - how else do you expect them to balance the output to white light
in the first place, given the different colour response of each channel?
This isn't a problem that is unique to scanning negatives - every colour
application has the same issue.
Anyway, it needs to adjust the exposure/stepping time of all three
scan lines to the one needed for the single most attenuated color
(which for CN film defeats most of the scan speed potential of a
tri-linear CCD).
Which is why, if you time the scanner, you will find it takes longer to
scan negatives than slides - even though slides have a higher Dmax, they
have a lower Dmin, especially compared to the blue channel of a colour
negative.
It'll also add cost and complexity to the sensor and supporting
circuits, but it is obviously possible.
No more cost or complexity than you need for the exposure control
anyway. What do you think those RGB gain controls are doing on your
scanner? You have told everyone on this forum often enough - why the
sudden doubt? ;-)
In my view it could also have negative impact on S/N (reduced fill
factor and potential well capacity). Larger silicon real estate
increases cost which seems a less likely avenue for a consumer scanner.
There is no reduced fill factor, well capacity or increased silicon real
estate required to implement individual exposure control. It doesn't
have any impact on the S/N because the device simply operates as per
normal - on completion of the exposure, the charge from all of the
elements in the line are transferred in parallel to a separate CCD for
readout. In some devices this is a single device, giving all three
channels on a single output, but generally there are 3 separate readouts
clocked in parallel. The charge is transferred through this CCD which,
unlike the imaging line, is blind to illumination to avoid image
smearing. In a linear CCD, the well size can be as big as you like,
just by making the channel wider, since there isn't a significant
constraint in that axis.
No.4 will increase noise for the weaker channel.
No more than Version 2, for the reasons I gave above.