OCR a scanned PDF
Drop in a scanned document and each page is read in this tab. You get its text to copy or save, and the same PDF with an invisible text layer, so search and selection work in any reader.
How to use it#
- Drop a PDF onto the box above, or click it to choose one. The first time, the recognition engine downloads, which takes a few seconds.
- Wait while each page is read. The box shows which page and how far through it the engine is.
- Download the searchable PDF, copy the text, or save it as a .txt file. Untick Mark where each page starts for one unbroken text.
What you get back#
You get two things back. The first is the text itself, in the box under the drop zone, to copy or save. The second is a searchable PDF: your original pages, untouched, with the recognised words laid over each one in invisible text. A reader shows the scan as it always looked, and search, selection, and copying find the words where they sit on the page.
That second one fixes the most common complaint about a scan: Ctrl+F or Find finds nothing, because the file holds pictures of words and no words. Once the text layer is there, search works in Acrobat, Preview, a browser, and a phone, and the search on your computer can find the file by its words.
Pages that already have text, such as a typed cover page in front of a scanned letter, are not read again. Their own text is kept, because it is exact and recognising it would only add mistakes. The results say which pages were read by OCR.
How accurate it is#
The engine is Tesseract, reading English. On a clean scan of printed text at 200 dpi or more it gets nearly every word right, and the results give its own confidence with what that means for checking. In testing, a scanned invoice at 200 dpi came back word for word, prices and pound signs included, at 95% confidence.
It struggles with handwriting, which it mostly cannot read, and with faint copies, photos taken at an angle, and text over pictures or coloured backgrounds. Tables come out as lines of words rather than rows and columns. Treat recognised text as a draft to check against the page, above all names, figures, and anything you will rely on.
Why the first page takes a moment#
Reading text from a picture needs a trained model, and this one runs entirely in your browser. The engine and its English model come to about 4.5 MB as they download, and 7 MB once unpacked. They load the first time you drop a file rather than with the page, and stay loaded while the page is open.
Each page is drawn at 300 dpi and read in the background, so the page stays usable while it works. On a recent laptop a page of typed text takes about two seconds, and a dense page or a slower phone takes longer. The timing under the results shows what your device took.
Frequently asked questions#
Is my PDF uploaded to be read?
#
No. The recognition engine is part of this page and runs on your device, so the document never leaves it. That matters for the kind of thing that gets scanned: contracts, medical letters, bank statements, and IDs.
How accurate is the text?
#
On a clean scan of printed English it is close to exact, and the confidence in the results says how sure the engine was. Below about 80%, expect misread words, and check names, numbers, and dates against the page before you rely on them. Handwriting is mostly beyond it.
Which languages does it read?
#
English only. Other languages written in the Latin alphabet partly work, but accented letters and words outside English are often misread. Each extra language would mean another model to download, so this page keeps to the one most visitors need.
Can it read a photo or a JPG of a document?
#
Not directly, because this page opens PDFs. Put the photos into one first with the image to PDF tool, which also turns phone photos the right way up, then drop that PDF here. A photo taken straight on, in good light, reads far better than one taken at an angle.
Can I get the text of a scan into Word?
#
Yes. Download the searchable PDF and drop it into the PDF to Word tool, which reads the recognised text and puts it into a .docx, one paragraph at a time. Or copy the text from the box here and paste it into Word or Google Docs. Either way the words come across and the fonts, columns, and tables of the original do not, so expect to tidy the layout.
Does the searchable PDF look any different?
#
No. The pages are the original ones, and the text layer is written in a mode readers lay out but never paint, so the file looks and prints exactly as before. It grows by a few kilobytes a page for the words.
My PDF already has text. Do I need this?
#
Probably not. If you can select words in it, the text is already there, and the PDF to text tool gets it out instantly. This page notices when every page already has text and says so.
Does it work on a password protected or copy protected PDF?
#
Only once the password is off. Remove it with the PDF password remover, then bring the unlocked copy back here. A file that opens freely but whose author forbade copying is refused, because reading its text out is the copying the author ruled out; its owner password lifts that lock in the same remover. A file locked only against changes can be read for its text, but a searchable copy needs that lock lifted with the owner password too.