On Fri, 11 Aug 2000, John Young wrote:
>Any of these could raise some obvious concerns. I'm curious if anyone
>might have a clear idea what "image matching software" is,
Yeah, I actually wrote some. About pi years ago, I was working
for a company whose main product was one of those "100 thousand
pieces of clip art" collections.
Anyway, they bought material from lots of different sources, but
they had a recurring problem which was that different people would
submit the same thing -- they'd grab the same out-of-copyright
publications and scan the illos for example, or contribute web
images that were just the same as another image but stretched ten
pixels wider or cropped ten pixels shorter. And of course, it
was in a dozen different formats -- gif, jpeg, tiff, aiff, and so
on. My former employer had no interest in spending money for
images it already had in its library.
They had another problem which was that some of their suppliers
would just take people's word for it that the images were not
under copyright to someone else, and were, um, occasionally
misled in ways that could have cost my employer huge amounts of
money if not discovered prior to publication.
My job was to create for them software that would recognize images
it had "seen before", even through most cropping, resizing, color
substitution, and format conversions, and which could find which
*one*, of a hundred thousand or more images, the new image reminded
it of. It was actually a very interesting problem, but I found a
good way to do it.
If anybody's interested in the algorithm, write me -- I was
quite proud of it. Has nothing to do with cryptography though.
I figure the FBI is interested in this for the same reason as
my former employer; they want a tool to track down image copyright
infringements.
Bear
