Scanned PDF to text: OCR in your browser
Recognise the words in scanned pages or photos of documents and copy them, without uploading the file.
Your file is processed in this browser tab and is not uploaded. Up to 100 MB and 1000 pages per file. Up to 100 pages per run; images up to 50 MB. On a typical computer a page takes a few seconds; the first run also loads the language data (about 1–3 MB per language).
This tool needs JavaScript. If it does not start, enable JavaScript in your browser and reload this page.
This browser cannot run the tool. Please use a current version of Chrome, Edge, Firefox or Safari.
This tool runs on desktop browsers only for now. Text recognition needs more memory than phones and tablets reliably have, and it has not been tested on them yet. Please use a computer.
No file selected yet.
Choose the document's languages, not the language of this page.
Thai: tone marks above upper vowels are often missed (ที่ → ที). Proofread the result.
Recognised text
How to use it
- Choose a scanned PDF, or a JPG or PNG image.
- Tick the languages of the document and, for a PDF, the pages.
- Select Recognise text. Copy the text or save it as a TXT file.
What it does
- Recognises printed text page by page and shows it under a page heading.
- Works with English, Simplified Chinese, Traditional Chinese, Japanese and Thai documents, and mixed-language pages when you tick several languages.
- Marks every page that could not be recognised, and after Cancel keeps the pages already done and lists the rest.
What it does not do
- It does not keep the layout: columns, tables and fonts become plain lines. To keep the page as it is and add searchable text, use Make PDF searchable.
- It does not read handwriting.
- Thai: tone marks above upper vowels are often missed (for example ที่ is read as ที). Proofread Thai text before using it.
- It does not run on phones and tablets yet. Use a computer with a current browser.
File limits
- Up to 100 MB and 1000 pages per file.
- Up to 100 pages per run; images up to 50 MB.
- On a typical computer a page takes a few seconds; the first run also loads the language data (about 1–3 MB per language).
Questions
- How accurate is it?
- It depends on the scan: sharp, straight, 300 dpi black-on-white pages give the best result; photos, low resolution and decorative fonts give more errors. Always check names and numbers.
- Which language should I tick?
- The language of the document, not of this website. Tick several for mixed pages; ticking unused languages makes recognition slower and can add errors.
- Is my document uploaded for recognition?
- No. Recognition runs in this browser tab with program and language files loaded from this site.
Privacy
- The file is read and processed by code running in this browser tab. It is not uploaded to our server or to any other service.
- The tool loads its own program files from this site; it does not contact other services while it works.
- Nothing is kept after you close the tab. Results exist only until you save them.