I use a simple script that uploads each file to chatgpt and collects the transcript. It works shockingly well compared to all ocr solutions.
Dne pet., 28. nov. 2025 ob 15:57 je Guy Fedorkow via cctalk < [email protected]> napisal/-a: > Greetings Restorers, > I think a number of us have wanted to restore software that's only > available as a scanned listing from a line printer. The original > printout probably wasn't the best typographic quality, and scanning > doesn't improve it. > As a first pass, OCR with tools like Adobe Acrobat can easily produce > a rough draft of the content in text form, but it takes almost as much > work to correct the many "typos" as it does to simply re-type the listing. > It seems like, with all this high-tech AI processing around, it > should be possible to take advantage of the limited character set, fixed > fonts, and restricted grammar that one might find in a listing to > resolve more of the ambiguities in character recognition. > Does anyone have an approach that's more efficient than generic OCR > and a long process of correcting typos on every line of code or comment? > Thanks > /guy > > >
