Digitising collections of microfiche - Re: Looking for opinions...

[Shaun Halstead]

[I'm intentionally not quoting anyone here]
Disclaimer:  I own a small company which provides microfilm and microfiche
scanning.

  A couple of problems anyone building a fiche scanner is going to run into
have already been addressed:  Moving film around to capture images and
resolution.  I use a Mekel M565 scanner.  It's roughly 3 feet long, 1.5
feet tall, and 2 feet deep, weighs around 80 pounds, and cost, used with
greyscale interface, $15k (ca 2007).  The length is to accommodate both the
internal control boards and the lenses required.  Image capture is
accomplished via a single-pixel-wide scan element, which is (from memory)
about a foot long.  The entire unit weighs around like 80 pounds.  While
running, it moves the desk it's sitting on.  It also uses (as I recall) a
65W lamp, which is also lensed, and all of the glass (lenses and film
carrier/platen) are photo-perfect, with zero defects.  The lamp itself is
interesting, as there are two versions available under the same ANSI
designation, one with a vertical filament, the other with a horizontal
filament.  Using the wrong filament orientation can cause some weird
artifacts to appear on scanned images, because of the high magnification.
I strongly suspect that an attempt using an LED source would face similar
(and possibly worse) issues.

  The system uses a dedicated interface card, with multiple FPGA's.
Typical scan times range from two to five seconds per frame, (including 0.5
to 1 second to reposition the platen).  The M565 has an auto-loader, though
I've seldom used it.  The auto-loader takes the form of an arm over the
platen, with a vacuum head and a solenoid to extend the head down to the
work table under the platen, pick up a fiche, hold it in place while the
platen is positioned and emptied (via pneumatic system), then drop the new
fiche onto the platen.  35mm images scanned at 300dpi equivalent are just a
hair over 40MP.  16mm images are, of course, smaller.

  A few issues that come to mind:
    How to detect where the frames are on the microfilm sheet.  Some fiche
have a very consistent grid layout, which is nice, because you can work out
the math then feed the grid information into the scanner software.  In my
experience, most fiche will have a grid layout, but not a consistent one.
There will be blank frames, skew, even frame size changes from column to
column or row to row.
    Film quality.  You're working with duplicate film, at least second,
probably 3rd, possibly even 4th generation, so quality was already an issue
20+ years ago, and it's an even bigger issue now.  All of the DEC film I
have (which I believe is already available) is diazo duplicates, which are
susceptible to fading over time, even when stored in proper conditions.
(In particular prolonged exposure to strong light, including sunlight and
the light from microfilm readers, accelerates fading of diazo films.)
    Light source.  Due to lensing requirements, LED's are probably out,
unless a way can be found to suitably diffuse or blend the source without
losing significant light.  This requires a very strong light source.

  In short, this is no small task.  I've been looking for newer
alternatives, but the resolution requirement is a killer on its own, and
then when you factor in other challenges like platen movement, frame
detection, light source, etc, it's very easy to be overwhelmed.  I would
love to see a viable alternative produced, but the efforts I've seen over
the years all had significant problems in terms of quality and/or speed of
operation.

--Shaun