I recently downloaded and tried VueScan with my old Nikon LS20 film
scanner. The software offers multi-pass scanning but can it really
reduce noise on this old scanner? I seems to function fine but was
wondering if it was any use since the slide makes multiple passes
through the mechanism.
The first time I used Vuescan was with a Nikon LS-20. I had been in the
process of writing some software to automatically align and merge some
individual scans for the sole purpose of noise reduction - in fact, it
is more effective in a scanner as old as the LS-20, with its noisy CCD
and 10-bit ADC, than it would be in some of the newer ones. The
multi-pass option of Vuescan offered to do that for me, so it was worth
evaluating. A few attempts showed that it wasn't going to deliver - at
least not with the LS-20 - but it also saved me a software task by
demonstrating a few unforeseen problems that meant the task was a lot
more difficult than I had initially envisioned. Consequently, I never
bothered with my own software solution.
The LS-20 scanner is different from many high performance film scanners
in that it moves the film past the scanner head rather than moving the
scan head and optics past the film. The result is that not only can the
position of each image change relative to others, but they can rotate
slightly in the drive as it is cycled back and forth and even scan at
slightly different steps due to minute slippage of the film in the
transport mechanism. So even if Vuescan included a facility to align
the centres of all of the images perfectly, they would be out of
alignment at the edges and corners. Of course, all the random structure
averages out, so the image still looks "cleaner", with less noise - but
it is much softer than the single pass scan because the averages are not
taken from the same points in the image for each pass.
What is needed is a proper image registration facility with rotation and
warping capability as well as basic sub-pixel alignment. Fortunately,
this is a common problem in lots of fields other than scanning,
including astronomy and medical tomography, so there are now quite a lot
of packages around that will do this correctly. For example, Astra
Image 2 has the facility to align several scans even when they are at
different scales and rotation. Mira is another application that does
the same thing. So there are lots of ways of doing this now - some
groups have even openly published C-code if you want to write your own
application.
In fact, there is a whole new branch of amateur astronomy developing
based on a modification of registered multiscanning, called "Lucky
Imaging", or LI. The biggest problem for a ground based astronomy
telescope is atmospheric turbulence, which blurs the images as the air
changes its density by minute amounts directly in the line of sight of
the telescope. However, every so often the atmosphere is relatively
still, so the image is quite sharp for just a fraction of a second.
Normally, the CCD on a telescope would be set to image for a long
exposure time, to capture the faintest amount of light and this means
that, just as with multipass scanning on your LS-20, the image is
blurred by the atmospheric turbulence. In LI they run the CCD quite
fast with short exposure times, capturing every frame. Each frame is
then analysed and the blurred images rejected automatically, leaving
only the sharp images to be "multiscanned" - after being properly
registered, of course. Some amateurs have produced planetary images
that rival those of the Hubble Telescope using LI, and more professional
arrangements have achieved the same on faint galactic objects.
Anyway, back to your LS-20 and Vuescan - yes multipass multiscanning
does reduce noise in the scanned image, and it works at least as well on
the LS-20 as in any modern scanner; however, the precision of the LS-20
isn't good enough for it to work without a noticeable image softening
due to mis-registration of the images. There are other applications
around that can compensate for these deficiencies though, so you can
overcome the problem if you really can't stretch to a modern scanner.