PDF Tools
OCR PDF
Read the words out of a scanned PDF and turn them into plain text you can copy, search or save. Recognition runs on your own device, so the document is never uploaded.
Text recognition happens in your browser, so results depend on scan quality and a long or high-resolution document may take several minutes. Keep the tab open while it works.
Your file and options
Choose a few pages at a time for the quickest results.
This tool runs entirely inside your browser. Your file is never uploaded, never stored, and disappears when you close the page.
Result
Choose a scanned PDF and the pages you want read.
About this calculator
This tool reads the text out of scanned pages inside a PDF, using text recognition that runs directly in your browser. It turns pages that are really just pictures of text into plain text you can copy or save as a .txt file. Nothing is uploaded, and how well it works depends heavily on how clean the original scan is.
How to use it
- 1.Choose the scanned PDF you want to read.
- 2.Pick the pages to process — a few at a time is fastest — and the language of the text.
- 3.Select Read text, then copy the result or download it as a .txt file.
Good to know
Accuracy depends on the scan. Straight, high-contrast pages of clear typed text read very well, while faint, skewed, low-resolution or handwritten pages produce mistakes.
Large files can take a while, because the recognition engine runs inside your browser on your own processor rather than on a server. The first page also has to load the engine, so it is always the slowest.
Pages that already contain real text are read directly, which is instant and perfectly accurate.
How it works in plain English
A scanned PDF is, from a computer's point of view, just a set of images stacked into pages, even though a person can read the words on them easily. This tool analyses each page image and identifies the shapes of individual characters and words, converting what it recognises into actual, selectable text rather than a picture of text.
Recognition quality depends heavily on what the scan looks like. Pages that were scanned at a decent resolution, sit straight rather than tilted, and use plain typed fonts with good contrast between the text and the background tend to come out very accurately. Pages that are blurry, skewed, faint, photographed at an angle, or full of handwriting are much harder to read reliably, and the extracted text will contain more mistakes that need checking afterwards.
Because this recognition work happens entirely on your own device rather than on a remote server, larger files or documents with many pages can take a noticeable amount of time to process, since your device's processor is doing all the work rather than a data centre. Once processing finishes, the recognised text is shown on the page where you can review it, and you can copy it directly or download it as a plain .txt file. The original PDF and the images extracted from it stay on your device throughout and are never sent anywhere or stored after you close the page.
Worked example
Someone has a ten-page scanned lease agreement, saved as a PDF from a photocopier years ago, and wants to search it for a specific clause about rent increases without rereading every page. They upload it to the tool and let it process each page in turn, watching a progress indicator since a ten-page document takes a little while on their laptop.
When it finishes, they get back plain text for all ten pages. Most of it reads perfectly since the original was cleanly typed and scanned straight, though a couple of faint, slightly crooked lines near a staple mark come out with small errors. They paste the text into a document, use its search function to jump straight to the rent clause, and manually fix the few garbled words nearby.
How scan conditions affect recognition accuracy
| Scan condition | Typical effect on accuracy |
|---|---|
| High resolution, straight, good contrast | Very accurate, few if any errors |
| Low resolution or blurry image | More misread characters and gaps |
| Page scanned at a skewed angle | Lines may be misread or dropped |
| Faint print or poor contrast | Text may be missed or garbled |
| Clean typed fonts | Reliably recognised in most cases |
| Handwriting | Much less reliable, expect heavy editing |
General behaviour of optical text recognition on scanned documents
Frequently asked questions
Will this perfectly recognise handwritten text?
Not reliably. Text recognition works best on clean, typed characters with consistent shapes, which handwriting rarely has. Neat, well-spaced handwriting might come through partially readable, but cursive or messy handwriting will likely produce a lot of errors. Expect to proofread and correct handwritten sections manually rather than trusting the output directly.
Why is processing slow on my document?
All the recognition work happens on your own device using your browser, rather than being sent to a powerful server to process. Documents with many pages, very large page images, or a slower device will naturally take longer, since your processor is doing every calculation itself. Closing other heavy browser tabs while it runs can help.
What format do I get the text in?
The recognised text appears directly on the page for you to review, and you can either copy it to your clipboard for pasting elsewhere or download it as a plain .txt file. It comes out as unformatted text, so original layout details like columns, tables, or fonts are not preserved in the output.
Does the scanned PDF get uploaded to check the text?
No. Recognition runs entirely inside your browser using JavaScript and WebAssembly, so the page images are analysed on your own device from start to finish. The PDF you select never leaves your computer, and once you close or refresh the page, both the file and the recognised text are gone.
How can I improve the accuracy of the result?
If you have any control over the source, a higher-resolution, straight, well-lit scan with strong contrast between text and background will always produce cleaner results. For scans you can't redo, reviewing the output carefully against the original pages, especially faint or crooked lines, is the most reliable way to catch and fix errors afterwards.