Allow me to add some more partially informed speculation about unreleased source code to the pile: Bill Stewart writes:
The description of Privasoft's method on their web page sounded like* it broke up a fax image into little squares and shuffled them around based on some proprietary (hence presumably weak) algorithm, somewhat like feeding them to a shredder or puzzle-box, then put them in a file you can ship with regular fax software or decrypt yourself. If that's the case, and if the squares are sufficiently big to have enough pixels for fax software to compress decently, then it's weak no matter _how_ strong the encryption algorithm that shuffles them is, because you can piece together matching edges like a jigsaw puzzle, or like a bunch of Iranian students with American Embassy shredded documents.
:} [...]
* The software was on the web page; if I'd had spare disk space that day and some slack time I could have downloaded the demo version and tried it, but this is what the description sounded like it meant.
The graphic encryption patents I found a couple of weeks ago (cited here then) describe a method that I don't believe would be susceptible to a "jigsaw" attack like this. The patented method specifies encrypting the image bitmap bit-by-bit (it suggests XORing with a (P)RNG stream). So the square granularity would fail your second criterion for the applicability of a jigsaw attack. However, I only have circumstantial evidence to suggest that the PrivaSoft software actually uses the patents I've seen. Perhaps PrivaSoft Inc. could comment on which patent(s) (if they have already issued) form the basis for the use of encryption in their product ? One last note: I would certainly have a hard time applying a jigsaw attack to the sample ciphertext image at http://www.privasoft.com/images.html (and enlarged at http://www.privasoft.com/scramima.html). Maybe it would be easier if the plaintext image consisted of a picture with more macroscopic structure. -Futplex <futplex@pseudonym.com>