Skip to content
UV pdfTool

OCR PDF

Make a scanned PDF searchable, with text you can copy, in 25 languages.

or drop files here

A document we wrote for this: nothing of yours is used.

Runs on your device. With Pro, you can choose our servers instead.

Quick facts

Takes
A scanned or photographed PDF. Pages that already hold text are skipped unless you choose “Every page”.
Gives
The same PDF with an invisible text layer over each page: searchable, selectable, and ready for the editor or for conversion to Word.
Languages
Twenty-five: English, Hindi, Bengali, Telugu, Marathi, Tamil, Gujarati, Kannada, Malayalam, Odia, Punjabi, Assamese, Urdu, Spanish, Portuguese, French, German, Italian, Russian, Arabic, Indonesian, Chinese (simplified and traditional), Japanese and Korean. Up to four can be chosen together for mixed pages. Words in Arabic, Urdu, Chinese, Japanese and Korean go into the text layer with a font that draws nothing, so they can be searched and copied like the rest.
Runs
In your browser, in a background worker, page by page at 300 dpi. The scan is never uploaded. The language data (0.3 to 2 MB per language) is downloaded from this site the first time and kept by your browser. With Pro you can choose OCRmyPDF on our servers instead, for scans of up to 2,000 pages; the file is deleted as soon as the result is fetched.
Plan
Free, with no account and no watermark: 10 tasks a day, on files up to 50 MB and 100 pages. Pro has no daily count and takes files up to 500 MB, with no page limit.

Measured on our scanned letters

We wrote a one-page letter in each of 18 languages, rendered each as a 200 dpi scan with a slight skew and paper grain, and had the tool read them. The figures below compare the recognized text with the words the letters were made from, character by character. The Indian, Latin and Cyrillic letters are set in the Noto fonts the editor uses; Arabic, Urdu, Chinese, Japanese and Korean, which those fonts do not draw, in the fonts the measuring computer has, as a printer would set them. Urdu is set in Naskh rather than the Nastaliq of most Urdu print, which reads less well.

Character accuracy of the recognition engine this page uses, on 2026-10-29 (build dd84427). A character counts as wrong when it is missing, extra or different.
SampleCharactersAccuracyEngine confidence
Hindi letter, printed, 200 dpi87798.6%90%
Telugu letter, printed, 200 dpi73699.6%91%
Bengali letter, printed, 200 dpi390100.0%93%
Marathi letter, printed, 200 dpi400100.0%95%
Gujarati letter, printed, 200 dpi38399.7%95%
Punjabi letter, printed, 200 dpi387100.0%94%
Odia letter, printed, 200 dpi36996.2%93%
Assamese letter, printed, 200 dpi38098.7%93%
Urdu letter, printed, 200 dpi35393.5%90%
Arabic letter, printed, 200 dpi34093.2%86%
French letter, printed, 200 dpi50599.6%95%
German letter, printed, 200 dpi486100.0%95%
Italian letter, printed, 200 dpi49099.8%95%
Russian letter, printed, 200 dpi431100.0%95%
Simplified Chinese letter, printed, 200 dpi12396.7%88%
Traditional Chinese letter, printed, 200 dpi12396.7%91%
Japanese letter, printed, 200 dpi15596.8%90%
Korean letter, printed, 200 dpi19899.5%93%

In practice this means one or two wrong characters per hundred on a clean printed page, usually in a conjunct or a mark, and up to 7 in the hardest script measured; a search for a word still finds it, and the editor corrects the rest. Low-resolution photos, faded copies and handwriting do far worse, and the page says so below instead of pretending otherwise.

What it looks like

The OCR PDF tool with the Hindi scanned letter loaded and Hindi ticked among the languages
Pick the languages the pages use. The first run downloads the language data from this site.
The result panel after recognition, with the searchable PDF ready to download and the editor offered as the next step
The result: the same pages, now searchable. Open it in the editor to correct a word, or convert it to Word.

What it cannot do

  • Read handwriting. The engine is trained on printed text; a handwritten note comes out as noise.
  • Make a poor scan good. Below about 150 dpi, or with heavy skew, shadows or a folded page, accuracy drops quickly. Scan to PDF straightens and sharpens a phone photo first, which helps more than any setting here.
  • Keep tables as tables. The words are recognized in reading order; the layout is not rebuilt. For a spreadsheet, recognize the text here and then use PDF to Excel.
  • Guess the language. A page in a language you did not tick is read with the wrong alphabet. Tick every language the document uses.
  • Change how the page looks. The text layer is invisible by design; the scan stays exactly as it was.

If something goes wrong

No page needed recognition, or no words could be read.
Either every page already has text (choose “Every page” to recognize anyway), or the scan is too faint or too small to read. Try a sharper scan, or run Scan to PDF on the photo first. Open Scan to PDF
The words come out in the wrong alphabet
The language of the page was not selected. Tick it and run again; for a bilingual page, tick both languages.
The first run is slow
The language data is being downloaded, once. After that, a clean page takes a few seconds on a laptop and longer on a phone. Keep the tab in the foreground while it works.
Your browser ran out of memory. Try fewer or smaller files.
Recognize fewer pages at a time: split the PDF, run OCR on the parts, then merge them again. Open Split PDF
A word is wrong in the result
Open the result in Edit PDF, where the recognized text is editable over the scan, and correct it. Open Edit PDF

About OCR PDF

OCR PDF recognizes the text in scanned pages and photographed documents and adds an invisible text layer, so the PDF becomes searchable, selectable and ready for editing or conversion. Recognition runs in your browser in many languages, including Indian scripts.

How to use OCR PDF

  1. Add the scanned PDF and choose the language or languages of the text.

  2. Recognition runs page by page on your device; the progress is shown.

  3. Download the searchable PDF, or open it in the editor to correct the text.

Frequently asked questions

Which languages are supported?
25 languages: English, Hindi, Bengali, Telugu, Marathi, Tamil, Gujarati, Kannada, Malayalam, Odia, Punjabi, Assamese, Urdu, Spanish, Portuguese, French, German, Italian, Russian, Arabic, Indonesian, Simplified Chinese, Traditional Chinese, Japanese and Korean. Up to four can be chosen together for mixed pages.
Does the scan's appearance change?
No. The page picture stays exactly as it was; the recognized text is placed invisibly on top of it.
How accurate is it?
On the clean scans measured on this page, about 99 characters in 100 come out right; low-resolution photos and handwriting do much worse. Scan to PDF straightens and sharpens photos first, which helps a lot.
Can I turn a scanned PDF into Word or Excel?
Yes. Recognize the text here first, then run PDF to Word or PDF to Excel on the searchable file.
Is the scan uploaded for recognition?
Not unless you choose our servers with Pro for a long scan. By default recognition runs in your browser; only the language data is downloaded from this site the first time, and the file stays on your device.

Guides