This is a simple utility to fix the OCR text found in a PDF file.
This project started as a helper utility to recover this file published by DOJ: https://www.justice.gov/epstein/files/DataSet%209/EFTA01012650.pdf
The PDF file contains two photos, taken with iPhone X 4 seconds apart (2018:12:18 18:54:31.409 and the other one 4 seconds later at 18:54:35).
I was able to fix 18 pages worth of the first image. There is still minor corruption remains, but the images comes along a photo of some cloth hangers with pieces of clothing hanging on them.
I think it is unlikely to be an important photo. It is probably photos of two items of clothing someone considering buying or wearing and sent to somebody for the advice.
Here is the result: 
(decoded with the utility and resaved to avoid being flagged for corrpted JPEG)
The expectation is that PDF file contains image and text, text in the PDF is base64 encoded image formatted to have 76 characters per line.
It stores the decoded files in current directory. As in page001.txt, page002.txt and so on.
Depends on pdfium library from Google.
Download from pdfium for your platform and place in the current folder. E.g. libpdfium.so for Linux.
cargo run -- ~/Downloads/EFTA01012650.pdf
Feel free to edit display_script.sh to ise whatever extra post processing you want to add.
My script just uses tr, base64 and display.
Screenshot: https://imgur.com/screenshot-gTnNrkW
- Ctrl+S: Save
- Ctrl+J: Jump to next I/l/1
- Ctrl+G: Jump to hex address
- Ctrl+D: Display image after the processing pipeline
- Ctrl+Space: Cycle next characters through common OCR mistakes (e.g., O/0, g/q, etc.)
- Ctrl+Enter: "Finalize" the current line. Adjust the number of spaces at the end of the line to align with the PDF (AI-generated logic)