The so-called "optical resolution" of flatbed scanners refers only to the
number of dots per inch that they scan. The actual optical resolution is
also a function of the spot size and shape,
In the case of this scanner it is the size of the CCD element that is
the issue, however expectations are set too high by quoting a "4800ppi"
specification.
Like most high resolution flatbed scanners, the Mikrotek uses an
oversampled CCD - where the pitch between adjacent elements is less than
the width of the element, in fact it is pretty close to half the width
of the element. This is achieved by manufacturing each linear CCD as
two rows of elements that are offset relative to each other by half a
pitch. Epson were the first to introduce this technology to commercial
scanners, calling it HyperCCD, Microtek call their version Sigma-Six
CCD, because 6 rows are required for the three colours. The approach
has been used on high performance military surveillance systems since
even before CCDs were invented!
The advantage is that the CCD is almost oversampled by itself, but
certainly is by the time that the lens MTF is included. Oversampling
ensures that the limiting resolution is less than half the sampling
density - which all sampling systems require to avoid aliasing. So
that's good in terms of the axis along the CCD - except for the fact
that to achieve zero MTF before the onset of aliasing, some MTF (ie.
resolved contrast) has to be thrown away below that hard Nyquist limit.
Because there is 4x as much collecting area per element as with a
traditional linear CCD though, the image stands up much better to
sharpening to recover this reduced MTF - indeed oversampling requires
post scan sharpening.
However, in the scanning direction the same resolution limit is imposed
- zero MTF at a resolution less than half the CCD's sampling density.
So, even though the stepper motor samples the data at twice the density
in that axis there just isn't any more resolution to be gained.
Furthermore, to create square pixels, scanning at 4800ppi requires the
samples in the CCD axis to be interpolated, which is a further loss in
MTF - so the overall resolution can actually be worse at 4800ppi than at
2400ppi. Though you probably need to undertake careful measurements to
discriminate this loss, there certainly isn't any resolution to be
gained.
This is, however, partially countered by the fact that you have twice as
many real samples, therefore getting the effect of 2x multiscan, and
hence 30% less noise.
I also suspect that
the stepper motor used to generate the scan motion induces vibration that is
not allowed to damp out before expossure of each line.
The vibration amplitude would have to be a significant fraction of 2x
the CCD pitch to be a problem, and I just can't see a step of half a
pitch inducing that level of vibration, even if it isn't allowed to damp
out.
In fact, many
flatbed scanners won't resolve more than 300-600 lines/inch.
By that I hope you mean line pairs, or cycles, since few flatbeds,
especially those specified at 2400ppi are quite that bad!
I have an Epson 1640SU with a transparency adapter. The adapter replaces
the usual lid of the scanner and includes a separate light source behind the
transparency. Using the adapter, I get perhaps 1000 lines/inch, although
I've never measured it.
I have, and it seems from that statement that you were not referring to
line pairs, making your earlier reference a long way out.
Limiting resolution of the 1640SU (MTF = noise) is 700cy/in, effectively
Nyquist for 1400ppi. However, this is the true resolution limit, when
nothing else is resolved, just like a camera lens resolution is defined.
Few people understand the difference between what such an image looks
like and what they typically expect a digital image to be, where the MTF
at Nyquist can be excessively high. In short, they mistake sharpness
for resolution. Anyone familiar with Photoshop should understand that
they can sharpen and image without changing its resolution - but few
seem to recognise the distinction.
At 450cy/in (900ppi Nyquist) the MTF of the Epson 1640SU is over 30% -
without sharpening - however sampling at 900ppi would result in
significant aliasing.
I have also used a Minaolta Dimage transparency
scanner, and the flatbed with the transparency adapter doesn't even approach
the quality of it..
But the Minolta aliases grain like few other scanners. :-(
