🚀 go-pugleaf

First, my apologies for a slightly shotgun approach, but I'm not sure where
to start.  Also, my apologies for including two Japanese newsgroups rather
than creating a separate Japanese version of this post.  Though my computer
and all software is Japanese, writing clearly in that language is still
quite difficult for me.

Background:  Image database of Japanese telephone cards.  Primary equipment
is Sony 505GX (w/external monitor) and Omron HS600 USB scanner.  All
hardware and software works well, though I'm quite likely overlooking some
of the features, especially in the Presto! scanning and photoediting
software that came with the scanner.  (For an embarrassing example, I didn't
know to look for the clipping tool until someone mentioned it in email...)

Immediate problem:  Producing quality scans.  Scanner uses tractor feeder,
and telephone cards are near the limits of its abilities.  I'm over 1,100
images already, but lots of rescans and fiddling are required.

Proposed solution:  Software to intelligently and quickly merge related
images.  Method would be to scan once in each direction, and then merge the
results.  Any recommendations?

Mid-range problem:  Identifying duplicate images.  Scans are slightly
inconsistent, and many cards have minor flaws.  In other cases, cards are
stock backgrounds with textual variations.

Proposed solution:  Generate sortable signatures based on color ratios of
.jpg images.  Sort index of signatures to find duplicates.  (Or possibly use
a n-dimensional vector and search for clusters in n-space?)  More concrete
suggestions or recommendations?

Long-term problem:  Searching from WWW.  Right now working on assumption of
no back-end support on the server side.

Possible solutions:  Client-side database or off-line searches.

Client-side database might be a Java application that downloads the
descriptive information to the client and prepares the searches there before
pulling the actual images.  For thousands of images a textual database is
okay, but even thumbnail images will probably be unfeasible.

Off-line searches would be easier, but not offer true interactivity.
However, I can do this with the tools I have already.  Descriptive database
would be searched and used to generate reports in HTML table format.  These
tables would actually use sized images to easily produce the thumbnails on
the fly from the local disk, and the table would lay them out nicely.  (I've
already done this kind of stuff for some other projects.)  The browser
images would be captured, and client-side maps would be prepared using the
collages of thumbnails.  From the web, a visitor would just select which
search results they wanted to see, and receive the collage image and
client-side map--clicking on the map would pull the large card image.
However, if the search wasn't available, they'd have to send me a request
and wait for me to prepare it.

Any comments, tips, useful URLs, etc. would be greatly appreciated.  I'd
appreciate an email copy, too, via my spam-filtered shanen@my-dejanews.com
e-address.  I'm going to attempt to monitor some of these newsgroups, but my
university's newsserver does not work perfectly reliably (to put it
diplomatically).

--
.a/ssig
And don't even think of spamming me.  I haf ze vays of dealing wiss ze
spamnuts.