1. 15
  1.  

  2. 4

    This is incredibly cool. It’s steganography, but applied in one of the most brilliant fashions I’ve ever seen.

    1. 3

      Seems like this would be trivial to detect, as fonts have a very well known and fixed representation that is present on… er, every computer in use. OCR the pages, overlay the letters, calculate deviance.

      I’d personally rather get a printed photograph where color levels were monkeyed with or other classic steganographic techniques were used. At least that way I’ve got some level of plausible deniability when state security came knocking.

      1. 3

        And you scan this and someone - say even an over eager document storage and delivery system - has the “OCR” feature on and thinks they are doing you a favor because they’re compressing your pdf …

        I’m also amused with this throwing of deep learning at everything. In this problem the model for the perturbations on the letters is known and therefore reverse model can be developed. Is this the new fashion: “Deep learning for matrix inversion”?

        1. 2

          I was always a fan of the microdot.

          1. 1

            Wouldn’t it be easier to have one font for zero and there second for one? Easy binary encoding?

            1. 1

              I’m honestly surprised intelligence agencies hadn’t already thought of this for barium meal traps. Or I suppose they have to pretend to be wowed.