|
Posted by =?Utf-8?B?TEFNUDkw?= on October 7, 2007, 7:59 pm
If you were Registered and logged in, you could reply and use other advanced thread options
You know? It occurs to me that both the font and the alphabet that MS
(Microsoft) uses in its product key is optimized for OCR (Optical Character
Recognition). Different fonts have different, say, "error clusters", sets of
letters that tend to be confused (or more exactly, with a higher probability
of OCR recognition error), like "O" and "0". Suppressing all but one member
of each "error cluster" from the alphabet would then lower OCR recognition
errors.
Since these error clusters vary from one font to another, the optimal
alphabet for each font would vary.
"Alex K. Angelopoulos (MVP)" wrote:
>
> > Thanks so much for your response, Alex. I do concurr with your idea of
> > readability and also avoiding "offensive" words in any Latin-alphabet
> > language. As for its random distribution, cryptographic theory says that
> > a
> > good cypher output is supposed to have such a random distribution.
>
> Yes. My reason for explicitly mentioning that was to drive home the fact
> that posting this information is going to have a negligible benefit for
> someone trying to crack keys.
>
> > It would be nice if there was already some written guidelines for this
> > subject. I still cannot find anything about it.
>
> I'm not too surprised about this, but that's because I spent some time
> talking about this issue with a friend who went to school in human/computer
> interactions. The ability to distinguish characters is a function of the
> font used and of how it is rendered - on screen or in print. On top of that,
> people from different regions also have certain expectations that influence
> legibility, even if using the same character set. That makes it hard to come
> up with any solid rules that don't change.
>
>
> > In the Engineering firm that I worked, we also had rules about not using
> > certain characters of the alphabet because of the possible confusion you
> > so
> > aptly pointed out. Those rules were pretty well established back in the
> > '50s; it is just that nobody was able to tell me where those rules came
> > from.
>
> I suspect they were 'common knowledge' developed over time and specifically
> oriented towards the style of handwritten block characters used by
> draftsmen. You've probably figured that yourself. ;)
>
> > So, suppressing certain characters seem to come from guidelines or best
> > practices coded long before cryptography-based product keys started to be
> > used by Microsoft.
>
> I hadn't thought about the historicity, but that makes sense as far as
> choosing the particular sets to avoid (other than the vowels). The kicker is
> the different typefaces issue.
>
> > Finally, I think I will stick to all digits and prune look-alike
> > characters.
> > Will leave the vowels alone for now.
>
> that theoretically means you could restore 1, 5, and maybe 0, and kill off
> the letter B. :)
>
> > P.S.: I thought MS would use a power-of-two character set for their
> > product
> > keys, like 32 characters, since it maps much easier into binary than 24
> > characters. I guess readability trumped over convenience!
>
> I found that odd as well, but then I realized that it probably doesn't
> matter much. We're not talking about a lot of data transmission and storage
> here.
>
>
| 
XML site map