PDF OCR — read the text out of a scan
Upload a scanned PDF or an image of a document and get back text you can copy and search. A PDF that already has a text layer is read directly, at no cost to accuracy — OCR only runs on the pages that actually need it.
Loading the tool…
Your file is uploaded to be processed, then handled under our retention policy.
How PDF OCR works here
- Choose a PDF or an image, and pick the language if you know it.
- The file is read directly if it already has a text layer — no OCR needed.
- Any page that comes back empty is recognised with OCR instead.
- Copy the text, or download it, once it is ready.
Extraction first, OCR only where it is needed
A born-digital PDF — one made by a word processor, not a scanner — already carries its text, and reading it out is instant and perfectly accurate. OCR is a fallback for the pages that have no text to read: a scanned page, a photographed document, a screenshot. Most real-world documents are not purely one or the other. A contract with one scanned signature page bound into an otherwise digital file gets extraction for the pages that have text and OCR for the one that does not, without you having to know or say which is which.
What OCR is actually doing
It is pattern recognition, not reading comprehension: the engine has learned what letter shapes look like and outputs its best guess for each one, run together into words and lines. That is why it fails in specific, recognisable ways rather than randomly — a low- contrast scan, an unusual font, or a skewed photograph all confuse the shape-matching in a way a human reader would not be confused by. Handwriting is not supported at all; it does not resemble the printed letterforms the engine was trained on.
Checking the result
OCR output reads fluently even when it is wrong — a misrecognised digit or a similar- looking letter swapped for the real one does not look like an error, it looks like a normal sentence with the wrong word in it. Names, numbers, dates and anything you plan to act on are worth checking against the original scan before you rely on them, the same caution this site gives about transcription for exactly the same reason.
Why this needs an upload
Most tools on this site run in your browser and only keep a copy on the side. This one does not — recognising text in an image is not something a browser can do on its own, so the file is genuinely sent to the server to be read. Nothing from it is saved to a searchable library the way the signed-in Documents feature does; it is handled once and put through the ordinary retention policy like any other upload.
Questions
What counts as a scan?
Any PDF or image where the words are pixels, not text — a photographed page, a document printed then scanned, a screenshot. You can tell by trying to select text in it: if nothing highlights, it is a scan.
Does this work on a PDF that already has real text?
Yes, and it skips the slow part when it can: a PDF's existing text layer is read directly, page by page, and OCR only runs on the pages that come back empty. A report with one scanned signature page bound into otherwise digital text gets both — extraction for the real pages, OCR for the one that needs it.
Which languages does it read?
English and Arabic, including right-to-left text. Pick the language before you upload if you know it; the wrong language produces confident-looking nonsense rather than an error, so it is worth checking the result against the original either way.
Do I need an account?
No — there is a per-file page limit without one, shown on this page, and a higher one once you sign in.
Is my document uploaded?
Yes — recognising text in a scan needs an OCR engine that does not run in a browser, so this is one of the few tools here that genuinely uploads your file rather than working on a copy kept locally. It is handled under our privacy page's retention policy, the same as any file sent to Scriptorium, and nothing is saved to a permanent library the way the signed-in Documents feature does.
What does it get wrong?
The same things any OCR engine gets wrong: unusual fonts, low-contrast scans, handwriting (not supported at all), and a skewed or badly lit photograph. It also cannot recover a photograph that was embedded in the original page — see PDF to JPG for that instead.
Turn what is said into text
Scriptorium's main job is transcription: bring a video or an audio file and get back text you can read, correct, search and export.