Bertram Nickolay, head of security technology at the Fraunhofer Institute for Production Systems and Design Technology (IPK) in Berlin, says that the heart of the reconstruction software that his team has spent years developing is powered by algorithms designed to recognize and process digital patterns and images.

The pieces of torn documents are scanned on both sides, and the digital images are then analysed by a cluster of 16 computers for 25 features, including colour, shape, texture, handwriting and typeface, Nickolay says. Just like a person doing a jigsaw, the computer then groups the images into clusters with similar features, and finally fits pieces in each cluster together. The software should get better with time, Nickolay notes. "It learns as it processes."

According to Nickolay, the project will start with ten bags of torn documents this coming June. Initial scanning will be done by hand by 20 team members, but the team is "looking for solutions to mechanize this process."

It turns out that science fiction writers are once again ahead of the curve. In his excellent (and Hugo-nominated) 2006 book Rainbows End, sf author and computer scientist Vernor Vinge attacks the problem of digitizing entire libraries with characteristic gusto. Meet the Navicloud Custom Debinder:

Ahead of him, everything was empty bookcases, skeletons. Robert went to the end of the aisle and walked toward the noise. The air was a fog of floating paper dust. In the fourth aisle, the space between the bookcases was filled with a pulsing fabric tube. The monster worm was brightly lit from within. At the other end, almost twenty feet away, was the worm's maw - the source of the noise... The raging maw was a "Navicloud custom debinder." The fabric tunnel that stretched out behind it was a "camera tunnel..." The shredded fragments of books and magazines flew down the tunnel like leaves in a tornado, twisting and tumbling. The inside of the fabric was stiched with thousands of tiny cameras. The shreds were being photographed again and again, from every angle and orientation...
(Read more about the Navicloud Custom Debinder)

If you are interested in the logistics of scanning millions of pages, read Encyclopedia Googleactica - Google To Put All Human Knowledge Online.

Update 11-May-2007: Thanks to reader RandomAction who wrote in with the same idea (I independently came up with it). RandomAction cited an article in spiegel.de:

Rights activists interrupted the project [by the East German Staatssicherheitsdienst to destroy the material] and rescued a total of 16,250 garbage bags full of scraps. But rescuing the history on those sheets of paper amounted to an absurdly difficult jigsaw puzzle. By 2000, no more than 323 sacks were legible again -- reconstructed by a team of 15 people working in Nuremburg -- leaving 15,927 to go. So the German government promised money to any group that could plausibly deal with the remaining tons of paper.
(From New Computer Program to Reassemble Shredded Stasi Files)

(End Update.)