Hello all,  This is my first posting to the MMD, although many of you
may know me from my activities in the Rollscanners discussion group.
First, a brief introduction:

I've been monitoring MMD for almost 18 months now.  I already know
many of you better than you can imagine from reading your postings
over the last six years.  (Yes, some of us actually _do_ scan the
Archives all the way back to Day One!)  Your experiences and advice
to each other (along with several books and nine solid months of my
own research) helped me to do my first complete player rebuild without
any significant destruction to piano or house (I don't know which would
have been worse!)

For the last 14 years, I have been involved in optical scanning
and image processing.  In my last position, I was Vice President
of Engineering and R&D for a small specialty scanner company.  Our
specialty was high-density two-dimensional barcode printing and
recognition.  As a result, scanning and image processing are things
that I'm more than a little familiar with.

Enough introduction -- on to Roll Scanning !

In a nutshell, various people in the Rollscanners group have combined
efforts to make high-quality, archival-quality roll scanning more
reachable and available to the average mechanical music enthusiast.

You can visit us to see what's happening at our discussion group:
http://groups.yahoo.com/group/rollscanners  or at our website
http://www.iammp.org/

The website has links to the discussion group and design files
associated with current and previous design efforts.

Spencer Chase has contributed his proven portable roll-scanning
transport design, which is being documented by Jack Breen for
duplication by others.  A number of small alterations will be made
to the design in this documentation process, primarily to select
components and materials that will provide equivalent performance,
but which are more readily available to and more easily assembled
by hobbyists.

I am contributing the scanner controller design, driver software
and *basic* scanning applications.  The application code is based on
work done by others in the group.  The driver software and feature
extraction code is my own design.

The hardware is compatible with pretty much any line image sensor you
can throw at it (see the IAMMP website for details -- I recommend the
RollSCAN-1 FAQ's as a starting point).  It will work with most line
imaging chips -- CCD's [charge-coupled devices] and contact image
sensors [CIS's] -- without external interface circuitry.

It supports front or backlighting, LED or CCFL (fluorescent tubes),
color or monochrome scanning, and a variety of resolutions and step
sizes.  Along with the built-in stepping motor drive circuitry, it is
pretty much a complete document scanner on a single board.

Many, but not all of the scanners will be built using 12" contact image
sensors that have been acquired by Bob Pinsker.

In answer to some of the software accuracy questions posed here
recently:

Every measurement is made to sub-pixel accuracy.  Even when scanning
at only 200 dpi, edge positions are estimated to closer than 0.002"
resolution.  Assuming the drive system doesn't introduce instability
into the measurements, internal measurement accuracy in the travel
direction is always much better than the scanner resolution.  This is
also true for hole position measurement across the roll.  The hole edge
positions are measured to an accuracy of at least 2-3 times better than
the scanner's pixel resolution.

Punch steps are back-calculated by determining punch geometry, then
by "differencing" as many different hole (slot) lengths as possible
(after measuring and eliminating bridging) to see if there is a strong
correlation around a particular step increment.  It is completely
unnecessary (and, in fact, undesirable) to alter the scanner's step
increment to match up with the punch step.  As long as the scanner's
resolution in the travel direction is at least three times the
punch-step increment there will never be a loss of accuracy.

The software permits measurement and cataloguing of multiple punch
sizes and shapes (e.g., small snake-bite holes, etc.) and makes
provision for offset and interleaved multi-row tracker scales
(as for some pipe organs, etc.).

Options are provided for storing the slot lengths either as real
measurements (full-resolution floating-point lengths) or in increments
of punch steps.  Storing real measurements lets you go back and
correlate multiple rolls of the same title to see if the estimates can
be improved.  (I have not written the application to do so yet, but the
storage format anticipates the need to do so).  "Real measurement" mode
also accommodates accurate measurement of "asynchronously punched"
rolls, even if there is no method currently available for re-punching
them.

Alternative output file formats will include: compressed roll image
files (very large) and MIDI file output.  Optional configuration files
will permit accurate MIDI recording and playback of non-chromatic
tracker scales.

Rather than try to explain everything here, we're probably better off
if you ask specific questions.  I will be happy to answer everything
I can, so fire away.

Thanks,
Gene Gerety

 [ Thanks very much for the news, Gene.  I'm quite impressed with the
 [ experts assembled and the momentum generated by Richard Stibbons
 [ and the Rollscanners group.  -- Robbie