Comment

Comment on Recreating uncensored Epstein PDFs from raw encoded attachments

<- View Parent

kescusay@lemmy.world ⁨1⁩ ⁨day⁩ ago

Long story short:

Some of the emails in the file dump had attachments.
The way attachments work in emails is that they’re converted to encoded text.
That encoded text was included - badly - in the file dump.
So it’s theoretically possible to convert them back to the original files, but it will take work to get the text back. Every character has to be exactly correct.

Source: I’m a software developer and I’m currently trying to recover one of these attachments.

source

Sort:hotnew top

proudblond@lemmy.world ⁨1⁩ ⁨day⁩ ago
Godspeed friend

source
apftwb@lemmy.world ⁨1⁩ ⁨day⁩ ago

I’m a software developer and I’m currently trying to recover one of these attachments.

🫡

source
apftwb@lemmy.world ⁨1⁩ ⁨day⁩ ago
Are you having as much trouble with OCR as the article author? I would have thought OCR was a solved problem in 2026 even with poor fonts.

source
- kescusay@lemmy.world ⁨1⁩ ⁨day⁩ ago
  I’m not having trouble with it as such, it’s just a slow and painstaking process. The source is crappy enough that an enormous number of characters need to be checked manually, and it’s ridiculously time-consuming.
  
  source
- floofloof@lemmy.ca ⁨1⁩ ⁨day⁩ ago
  I wonder if they gave considered crowdsourcing this, having many people type in small chunks of the data by hand, doing their own character recognition? Get enough people in and enough overlap and the process would have some built-in error correction.
  
  source
  - apftwb@lemmy.world ⁨1⁩ ⁨day⁩ ago
    I mean the problem is that even with human eyes it’s still really hard to tell l and 1 in that font.
    
    source
    Kevlar21@piefed.social ⁨1⁩ ⁨day⁩ ago
    Not an expert at all but I’m genuinely curious how long it would take to check all possibilities for each I or 1? Is that the full length of the hash or whatever? So in this example image we have 2^8 =256 different possibilities to check? Seems like that would be easy enough for a computer.
    
    source
- Taldan@lemmy.world ⁨18⁩ ⁨hours⁩ ago
  OCR is mostly good enough. Problem here is we have 76 pages that we need to be read perfectly, with a low fidelity input
  
  We also have very little in the way of error correction, since it’s mostly not human readable
  
  source
trolololol@lemmy.world ⁨1⁩ ⁨day⁩ ago
Curious here, this is base 64? And what’s behind it is more often than not an image or text? And you need to do ocr to get the characters?

Maybe for the text it could use a dictionary to rubber stamp whether that zero is actually a letter oh, etc etc?

I’m curious to know what the challenge is and what your approach is.

source
- kescusay@lemmy.world ⁨1⁩ ⁨day⁩ ago
  Yes, it’s base64. And what’s behind it could be anything that can be attached to an email.
  
  In this case, it’s a PDF. If the base64 text can be extracted accurately, then the PDF that was attached to the email can be recreated.
  
  The challenge is basically twofold:
  
  There’s a lot of text, and it needs to be extracted perfectly. Even one character being wrong corrupts it and makes it impossible to decode.
  
  As the article points out, there are lots of visual problems with the encoded text, including the shitty font it’s displayed with, which makes automating the extraction damn near impossible. OCR is very good these days, but this is kind of a perfect example of text that it has trouble with.
  
  As for my approach, I’m basically just slowly and painstakingly running several OCR tools on small bits at a time, merging the resulting outputs, and doing my best to correct mistakes manually.
  
  source
  - trolololol@lemmy.world ⁨23⁩ ⁨hours⁩ ago
    Ah yes pdf is a clusterfuck where anything is valid I think, so minimal redundancy.
    
    Text and image formats are way more lenient and are full of redundancies.
    
    source