4 ms·

In some cases, coloring over parts of the image might still not be enough. Specifically, when all of the following are true: (i) the domain of possible entries

by nathell 12y ago

In some cases, coloring over parts of the image might still not be enough. Specifically, when all of the following are true: (i) the domain of possible entries is reasonably small (e.g., a number or a name and surname), (ii) the text is printed in a proportional (not fixed-width) typeface, and (iii) enough of the rest of the line is visible to infer the font size and kerning settings.

userbinator 12y ago
Actually, using fixed-width would be even easier since it reveals the length.
- bryondowd 12y ago
  I'd be curious to see some numbers backing one or the other, but I suspect proportional would be better. Fixed width narrows to a number of characters. So it depends how many possible values have x characters. Proportional narrows to a particular width, which could be more granular than character count. So it depends how many possible values turn out to be exactly x pixels across given the type settings.
  - cruise02 12y ago
    It doesn't really matter how many values turn out to be exactly x pixels across given the type settings, you still have to do an exhaustive search on different numbers of characters to find them. That problem is simplified by a fixed width font.
    - mokus 12y ago
      In general, though, a proportional face will give a larger number of possible widths, meaning more information to work with (fewer possible strings will map to any particular width). Additionally, depending on the type of information you are trying to recover, you may already know the number of characters (for example, a US SSN), in which case fixed-width typefaces mean you get no information you didn't already have, whereas a proportional one would at least give you a little bit of info. But on the other hard I would also guess fixed-length strings such as an SSN are more likely to appear in a more "table"-like part of a document than in the middle of a paragraph of text. I suspect it'd still be a bloody hard analysis either way, but at least according to this back-of-the-envelope theory a proportional typeface should preserve more information.
      - cruise02 12y ago
        Oh yeah, in the case where you already know the number of characters, a proportional case font definitely gives you more information. That's a good point. I was thinking more of the general case where any number of characters might be redacted from a document.
  - dalke 12y ago
    Proportional is easier to try to decode. See http://cryptome.org/cia-decrypt.htm http://cryptome.org/cia-decrypt.htm for an example from 2004 about inferring words from a redacted US document. From the Le Monde article: The space occupied by an "I" differs from that taken by "W," which can give additional clues, compared to the text-spacing known as "monospace," like that of often used in e-mailwhere all the letters have the same spacing. From the NYT article: In January, the State Department required that its documents use a more modern font, Times New Roman, instead of Courier, Mr. Naccache said. Because Courier is a monospace font, in which all letters are of the same width, it is harder to decipher with the computer technique. There is no indication that the State Department knew that.
ColinDabritz 12y ago
I think this form of analysis is amazing. It breaks the "The best solution was to actually cut out the parts of the page you don't want seen and then scan in the result." approach from elsewhere in this thread as well. Even cutting out each portion of each line isn't enough because you can reconstruct line sizes. You could cut out each individual word and string them together, with a single size "redaction" block perhaps, but that's a lot of work.
- tetrep 12y ago
  >You could cut out each individual word and string them together, with a single size "redaction" block perhaps, but that's a lot of work. Not really. OCR is plenty good enough to scan a document and then you can just replace anything you wanted redacted with a common string: [REDACTED] Even if OCR wasn't an option, it would still (probably) be easier to type up the document than it would be to physically cut out all the things you want to redact.

by nathell 12y ago

userbinator 12y ago

Actually, using fixed-width would be even easier since it reveals the length.

bryondowd 12y ago
I'd be curious to see some numbers backing one or the other, but I suspect proportional would be better. Fixed width narrows to a number of characters. So it depends how many possible values have x characters. Proportional narrows to a particular width, which could be more granular than character count. So it depends how many possible values turn out to be exactly x pixels across given the type settings.
- cruise02 12y ago
  It doesn't really matter how many values turn out to be exactly x pixels across given the type settings, you still have to do an exhaustive search on different numbers of characters to find them. That problem is simplified by a fixed width font.
  - mokus 12y ago
    In general, though, a proportional face will give a larger number of possible widths, meaning more information to work with (fewer possible strings will map to any particular width). Additionally, depending on the type of information you are trying to recover, you may already know the number of characters (for example, a US SSN), in which case fixed-width typefaces mean you get no information you didn't already have, whereas a proportional one would at least give you a little bit of info. But on the other hard I would also guess fixed-length strings such as an SSN are more likely to appear in a more "table"-like part of a document than in the middle of a paragraph of text. I suspect it'd still be a bloody hard analysis either way, but at least according to this back-of-the-envelope theory a proportional typeface should preserve more information.
    - cruise02 12y ago
      Oh yeah, in the case where you already know the number of characters, a proportional case font definitely gives you more information. That's a good point. I was thinking more of the general case where any number of characters might be redacted from a document.
- dalke 12y ago
  Proportional is easier to try to decode. See http://cryptome.org/cia-decrypt.htm http://cryptome.org/cia-decrypt.htm for an example from 2004 about inferring words from a redacted US document. From the Le Monde article: The space occupied by an "I" differs from that taken by "W," which can give additional clues, compared to the text-spacing known as "monospace," like that of often used in e-mailwhere all the letters have the same spacing. From the NYT article: In January, the State Department required that its documents use a more modern font, Times New Roman, instead of Courier, Mr. Naccache said. Because Courier is a monospace font, in which all letters are of the same width, it is harder to decipher with the computer technique. There is no indication that the State Department knew that.

ColinDabritz 12y ago

I think this form of analysis is amazing. It breaks the "The best solution was to actually cut out the parts of the page you don't want seen and then scan in the result." approach from elsewhere in this thread as well. Even cutting out each portion of each line isn't enough because you can reconstruct line sizes. You could cut out each individual word and string them together, with a single size "redaction" block perhaps, but that's a lot of work.

tetrep 12y ago
>You could cut out each individual word and string them together, with a single size "redaction" block perhaps, but that's a lot of work. Not really. OCR is plenty good enough to scan a document and then you can just replace anything you wanted redacted with a common string: [REDACTED] Even if OCR wasn't an option, it would still (probably) be easier to type up the document than it would be to physically cut out all the things you want to redact.