And you thought double shredding would save your privacy... http://unisci.com/stories/20012/0621015.htm I recently got their paper and although it is very interesting for image classification, it seems of limited use for reconstruction of printed text from shredded documents. The authors describe a method for classification of textures to a compact representation (a 2NK long vector) and a way of determining the most likely texture class a given

On Thu, Jun 21, 2001 at 07:08:14AM -0500, Jim Choate wrote: pixel is part of (by examining the pixels neighbours), with an error probability of 22% for 10 different textures. Although not described in the paper, I assume the pixels of the same class can be grouped into "patches" of the same texture, allow jigsaw-like problem solving. I don't think this texture classification method is going to work on normal printed text, as it has no real distinguishing texture. For shredded pictures the classification method will probably be effective in reducing the pictures of the shreds to smaller distinguishing values that can be used for a jigsaw solving algorithm. So as I see it, the real question is: how viable is the reconstruction of the complete "puzzle" given most/all of the pieces, some valueset indicating the probability two pieces are neigbours and maybe some information on the overall picture? Looks like a hard problem. With kind regards, Wouter Slegers