PDF OCR: get the text out of scanned PDFs
A scanned contract, a photographed form or a PDF exported from a phone scanning app looks like text, but to a computer it is only a picture: you cannot copy a sentence, search for a name or paste a table into a spreadsheet. This PDF OCR tool reads those pictures with optical character recognition and gives you the words back as plain, editable text.
Each page is rendered at scanning resolution (300 DPI by default) and read by Tesseract, the open-source OCR engine, running inside your browser as WebAssembly. Pages that already contain real text are detected and copied directly, so a mixed document — typed pages with a few scanned annexures — comes out complete and fast. Next to the text you also get a searchable PDF: your original pages with an invisible text layer on top.
It works in 30 languages, including Hindi, Marathi, Bengali, Tamil, Telugu, Gujarati, Urdu, Arabic, Russian, Chinese and Japanese, and can read two languages on the same page. Nothing is uploaded, which matters for bank statements, ID copies and medical letters.
How to OCR a scanned PDF
- Drop your PDF onto the upload area or click it to choose the file.
- Pick the Document language. For pages that mix two languages, such as Hindi with English, also choose a Second language.
- Keep the scan resolution at 300 DPI, or choose 400 DPI for tiny print. To save time on long files, type the pages you need in Pages to process, for example
1-3, 8. - Press Start OCR and follow the progress for each page. The first run downloads the language data, a few megabytes that are then cached.
- Proofread the text in the box, then use Copy text, Download .txt or Download searchable PDF.
PDF OCR features
Real 300 DPI rendering
Pages are drawn at true scanning resolution before recognition, the setting OCR engines are tuned for, rather than at screen size.
Mixed documents handled
Pages with a genuine text layer are not OCR’d again; their exact text is copied into the result, so nothing is misread or doubled.
30 languages, two at once
Latin, Devanagari, Bengali, Tamil, Telugu, Kannada, Malayalam, Gujarati, Gurmukhi, Arabic, Cyrillic, Thai, Chinese, Japanese and Korean scripts.
Page-by-page progress
Each page shows whether it was read, skipped or empty, with Tesseract’s confidence score, and you can stop at any time and keep what is finished.
Text and a searchable PDF
Download the words as a .txt file or get your original PDF back with an invisible, selectable text layer added.
Private by design
Rendering, recognition and PDF writing all happen on your device. Close the tab and nothing remains.
What OCR can read well — and what it can’t
OCR is very accurate on printed text with good contrast: office letters, invoices, books, typed forms and statements usually come out 95–99% correct. Accuracy drops with blurred phone photos, faint photocopies, coloured or patterned backgrounds, decorative fonts, very small print and handwriting, which Tesseract is not trained to read.
The result is plain text in reading order. Bold, fonts, columns and table borders are not rebuilt, so a two-column page reads column by column and table cells become lines separated by spaces. Paste the text into a document or spreadsheet and tidy it from there. If your PDF already has selectable text, PDF to Text extracts it instantly and exactly, without OCR.
Getting the cleanest text
- Choose the right language. Reading a Hindi page with only English selected produces nonsense. Add the second language for bilingual forms.
- Use upright pages. Sideways or upside-down scans read badly. Fix them first with Rotate PDF.
- Raise the resolution for small print. 400 DPI helps with footnotes and fine print; 200 DPI is enough for large, clear type and is faster.
- Scan, don’t snap. If you are capturing paper with a phone, Scan to PDF straightens the page and boosts contrast, which gives OCR a much easier job.
- Check numbers and names. Even at high confidence, OCR can confuse 0 and O, 1 and l, or 5 and S. Proofread amounts, dates and IDs before relying on them.
Resolution, speed and confidence
| Setting or score | What it means |
|---|---|
| 200 DPI | Fastest; fine for large, clean print such as typed letters |
| 300 DPI (default) | The standard for OCR; best balance of accuracy and speed |
| 400 DPI | Slower, uses more memory; helps with very small or thin text |
| Confidence 80–100% | Clean result — a quick proofread is enough |
| Confidence 60–79% | Expect some wrong words; check names and figures |
| Confidence below 60% | Poor scan, wrong language or handwriting — improve the source and run again |
PDF OCR FAQ
How do I extract text from a scanned PDF for free?
Open the PDF on this page, choose the language of the document and press “Start OCR”. When the pages are read, copy the text or download it as a .txt file. There is no sign-up, watermark or page limit.
Is it safe to OCR confidential PDFs here?
Yes. Your file never leaves your browser: the pages are rendered and recognised on your own device. Only the OCR engine and language files are downloaded from a public CDN, and nothing about your document is sent back.
Can it OCR Hindi and other Indian languages?
Yes. Hindi, Marathi, Nepali, Bengali, Tamil, Telugu, Gujarati, Kannada, Malayalam, Punjabi and Urdu are supported. For documents that mix an Indian language with English, select both languages.
How long does PDF OCR take?
Usually 3–10 seconds per page on a modern laptop and a little longer on phones, plus a one-time download of the language data. Use “Pages to process” to read only the pages you need from a long document.
Does PDF OCR keep the formatting and tables?
No. OCR returns the words in reading order as plain text, so fonts, columns and table layout are not reproduced. The searchable PDF download keeps the original look and simply makes the words selectable.
Why is some of the text wrong or missing?
Common causes are the wrong language, a blurred or skewed scan, very small print or handwriting. Select the correct language, try 400 DPI, make sure the pages are upright, and check the confidence score shown for each page.