Polar said:
At a photography equipment shop, the assistant recommended one of the
'photo' flatbed scanners (Epson & Cannon) instead of a dedicated 35mm film
scanner. They range in resolution from 2400 to 4800 dpi, i.e. similar to the
dedicated scanners, at lower prices, plus they also scan prints. One of them
is A3+, another takes film up to 120. Is there a catch?
Yes, the catch is that resolution is not the only measure of scanner
performance. Another measure of performance is MTF, or modulation
transfer function. This is just a measure of how much contrast the
scanner can reproduce at different resolutions. It is just the
scanner's equivalent of the frequency response plot that you used to see
on the back of quality audio cassette sleeves. Assessing scanner
performance requires consideration of resolution and MTF together, as
well as other parameters such as density and dynamic ranges.
Like most technologies, there is a "god law" that sampled data cannot
unambiguously reproduce signals with more than half the sampling
resolution. So if you have a scanner that is, say, 4000ppi then it
cannot reproduce signals which are more than 2000cy/in, or 2000lp/in. In
other words, you need at least two pixels per cycle - crudely one for
the positive half and one for the negative half. If you sample signals
which have more resolution then this results in aliasing, or beat
patterns. I am sure you have seen these before, if not in scanned
images then on TV with presenters wearing fine patterned jackets or fine
striped shirts. In traditional sampled systems, such as digital audio,
the signal is deliberately filtered to less than this limit before being
sampled so as to avoid audio beats and aliases.
This is effectively what most flatbed designs try to achieve, by using a
staggered CCD array. This is a linear CCD where each colour is actually
produced from two separate lines of sensors, offset from each other by
half a pixel pitch. We can go into the technical details of how this
works if you like, but the key point is that this approach produces an
MTF which is zero exactly at half the sampling resolution and pretty
small above that. The consequence is that the scanner can reproduce all
of the information presented to it without aliasing, right up to the
maximum resolution quoted.
However, in the audio the signal is electronic and it can be restricted
to the unambiguous band by steep cut-off electronic filters. In optics
there is no practical equivalent. Consequently, by ensuring that no
contrast is present in the sampled scanner image at the limit where
ambiguity begins, there must also be a reduced contrast in the region
where the scanner image can be reproduced unambiguously. The
consequence of this reduced contrast, or low MTF, is that the image
looks soft - even though it technically has the full resolution quoted,
the MTF at the limit of that resolution is zero and it is also low for
quite a way up to that limit.
In a dedicated desktop film scanner the emphasis in the design is to get
the maximum MTF right up to the limit of resolution even if some
significant MTF is present beyond that limit. Consequently the scanned
images have more contrast in the detail and look sharper - even though
the limit of resolution may technically be the same as the flatbed,
there is essentially no MTF reduction below that limit.
You might ask why these scanner types differ, and the answer is the
usual combination of compromise and cost. In a flatbed scanner, the
device is not just used for film, and some of the media may have very
specific spatial frequencies that extend over large areas of the subject
and these will alias into particularly obvious and objectionable
patterns. Printer dot patterns are just one example. So the flatbed
scanner design is generally optimised to eliminate these effects by
ensuring that it is physically impossible to create aliasing at the
limit of resolution. If no aliasing is present at the maximum
resolution then it can safely be downsampled by algorithms that minimise
its introduction. Such algorithms all contain the digital equivalent of
the sharp analogue cut-off filter used in the audio signal.
On the other hand, the finest detail that is usually presented to the
dedicated film scanner is the film grain, which is a random range of
spatial frequencies, few of which correlate between samples.
Consequently, any aliasing that occurs in the dedicated film scanner is
usually only an amplification of the random grain structure, there is no
beat pattern or typical alias image. The original grain may be finer
than a pixel, but it is all reproduced as coarse as a pixel but with the
original contrast. In some cases, with very high resolution film and
camera lenses some image aliasing can occur, although it is generally
quite small with current 4000 and 5400ppi scanners.
Ideally, dedicated film scanners should use the staggered CCD approach
as well to ensure that this simply does not happen and to eliminate
grain aliasing entirely - but there is a cost. That would mean that
your film scanner would have 4x as much data to produce and a more
expensive CCD - for a very limited set of images from cameras and film
which are capable of producing the problem. Well, one day it might
happen, but it isn't there yet.
If you have your images scanned with a proper drum scanner then the
operator will have a control called the aperture (or similar) which
basically adjusts the sensor pixel size independently of the resolution
of the scan - effectively adjusting the amount of overlap between pixels
somewhere between that of the desktop film scanner (zero) and the
flatbed (up to 50%). This allows the drum scanner to obtain the best
tradeoff between MTF (ie. apparent image sharpness) and aliasing for
each image and film type. Unfortunately, this is not a facility that is
available with any current dedicated desktop scanners.
So the bottom line is that a flatbed may have the same resolution on
paper as a film scanner, but the image from the dedicated film scanner
will be considerably sharper, at the expense of some visually amplified
grain and possible image aliasing if your film, lens and camera
technique are absolutely tip top.
Is the quality
comparable to that of dedicated filmscanners?
No. They are different trade-offs.
I have a large amount of 35mm
slides plus a small number of negs to scan & I'd like to be able to print
them on special A4 photo papers. Are flatbeds really a good alternative?
If you restrict your output to A4, then the best of the current crop of
flatbeds scanning at 4800ppi will do an adequate job, even with
reasonable cropping. You simply won't see the reduced MTF and the
elimination of aliasing will produce nice smooth images. However, like
most things, once you get used to printing at that size you will want a
little bit more, or there might be situations where a bit more cropping
is necessary and then the MTF limitations of your flatbed will start to
be objectionable. Ultimately, you want the MTF of the film to be the
limit on your print size, not the MTF of the tool that you use to
digitise those film images. It is essentially the same argument for
choosing the best lenses for your camera.
