this post was submitted on 20 Aug 2023
1097 points (97.6% liked)

Memes

45887 readers
1737 users here now

Rules:

  1. Be civil and nice.
  2. Try not to excessively repost, as a rule of thumb, wait at least 2 months to do it if you have to.

founded 5 years ago
MODERATORS
 
you are viewing a single comment's thread
view the rest of the comments
[–] JohnDClay@sh.itjust.works 4 points 1 year ago (1 children)

But the letter shapes change slightly depending on what letters are before and after.

[–] Perfide@reddthat.com 3 points 1 year ago (2 children)

You don't just scan individual letters, you also scan a bunch of different combos of letters next to each other, as needed. For example, you're gonna want specific scans for things like "ea","ee", "eu".

[–] JohnDClay@sh.itjust.works 2 points 1 year ago (1 children)

Getting several examples of every letter combination gets very hard very fast. Just lowercase, to get 5 examples of the letters before and after each letter is nearly 100k examples. You'd probably be better off doing some machine leaning shenanigans to simplify the process from training data.

[–] Perfide@reddthat.com 3 points 1 year ago* (last edited 1 year ago)

I didn't say every letter combination. I said the ones you need. Letter combos that do not connect to each other aren't needed. Still though, you're right that machine learning is needed... the good news is it's already been done before, and the code is open source. StuffMadeHere on youtube already built a fully functional prototype that impressed if maybe didn't fool forgery experts. https://youtu.be/cQO2XTP7QDw