Printers betray document secrets
jerrold.leichter at smarts.com
Thu Oct 21 14:09:19 EDT 2004
| It turns out that their techniques aren't all that useful.
| Changing laser printer cartridges changes the results.
| You might find that two documents were printed
| by the same printer, but it doesn't give you the
| options for tracking it down that manual typewriters did.
Actually, they say they can identify the make and model - which is about all
you could do with a typewriter. Going further, in either case, means tying a
particular piece of text to a particular writing instrument to which you have
Changing printer cartridges will certainly work, but then again simply replac-
ing the typewriter will, too. Any identification of physical objects can only
work as long as the physical object isn't replaced. In practice, there's a
great deal of inertia in replacing physical objects, for cost, convenience, and
other reasons. So such identifications may still be useful.
| And the differences don't identify a specific printer
| in a way that can be tracked, e.g. identifying a serial number
| that could be looked up from warranty records.
A bullet can't be tied to a gun's serial number, but that doesn't make it
useless to examine bullets.
| It's not clear that they work at all with inkjet printers,
| and changing ink cartridges is even more common than
| changing laser printer cartridges.
The technique is based on variations in dot pattern that ultimately come down
to small variations in mechanical parts, usually the gears that drive the
paper. Laser printer cartridges are deliberately designed so that (just
about) all moving/wearing parts are part of the cartridge. So most variations
in the results are necessarily tied to the cartridge. That's not true for ink
jets. While the paper describing all this isn't yet available, from what is
published I don't think they are making any claims about inkjets, just laser
printers. However, they seem to believe the same general approach - look for
variations due to variations in manufacture that don't produce artifacts that
are visible to the naked eye, so don't need to be and hence are not controlled
- would work. Whether the source of the variation would be in the ink
cartridge or in the fixed mechanicals, who can say at this point.
| If you're sloppy,
| you've probably got a bunch of partly-used cartridges around,
| so even if you want to print out a bunch of ransom notes
| or whatever, you don't even have to go to Kinko's
| to get them to be different.
| If printer makers want to build in watermarking to
| make everything they print traceable, the way many of them
| check for documents that look like money and don't print them,
| they could hide patterns that survive cartridge changes
| (would you notice a few inverted pixels on a 600x600dpi printout?)
Actually, this would probably be noticable in certain pictures. But slight
variations in pixel spacing - which is what these guys look for - is not
visible. (In fact, the origin of this work seems to have been work in the
opposite direction: Early laser printers had a problem with banding, due to
periodic variations in paper movement causing variations in pixel spacing.
The trick was to find out how much variation you could allow without visible
artifacts and then get to that level cheaply. But there is still plenty of
variation left for appropriate software to find.) You could probably play
games with pixel sizes, too.
| But even then, inkjet printers are dirt cheap;
| when they're on sale, they're essentially a free enclosure
| in a box of overpriced printer cartridges,
| so even of the printer wants to rat out the user and
| it's not easy to change the serial number PROM,
| you can just replace the printer.
One could say the same about most physical objects that end up being used
for identification. You would think that fibers would be useless for
identification, for example - you can always throw out the clothing you were
wearing and buy a new tee shirt. Still ... the real world has a great deal
The Cryptography Mailing List
Unsubscribe by sending "unsubscribe cryptography" to majordomo at metzdowd.com
More information about the cryptography