Files tool

PDF OCR & Scanned Text Extractor

Extract text already inside a PDF, or run OCR on scanned pages in your browser and save the recognized text.

Runs in your browser. No upload.

Text-based PDFs can be extracted without OCR.

About this tool

PDF OCR & Scanned Text Extractor made simple.

Scanned-page OCR loads a locally hosted language model only when you start it; the model is cached by your browser. OCR can misread handwriting, small print, columns and poor scans, so check the result. This tool exports text and does not rebuild a searchable PDF.

Guide

Using PDF OCR & Scanned Text Extractor

This page pulls the text out of a PDF. If a page already holds selectable text, the page copies it directly. If a page is only a picture, as in a scan, the page runs optical character recognition, or OCR, to read the letters from the image. You choose the language of the printed text and the pages to read, press one button, then copy the text. The PDF is opened and read on your own device. The result is plain text only. The page does not build a new searchable PDF, keep the layout, or read handwriting well.

How to use it

  1. Press the file box under Choose a scanned PDF and pick a PDF from your device.
  2. Open Text language and choose the language printed on the pages. Only one language can be used in each run, and it matters only for pages that are pictures.
  3. In Pages to scan, leave the word all, or type pages such as 1-3,5. Numbers must be between 1 and the last page of the file.
  4. Press Extract text / OCR. A progress bar appears and the status line counts the pages, for example Reading page 1 of 3. A scanned page needs a language file to download first, so the first run can take longer.
  5. Read the Extracted text box. Each page begins with a line like --- Page 1 ---, and pages are separated by a blank line.
  6. Press Copy text to paste it into your own editor, or press Download TXT to save the text as a plain file. Press Clear to start over.

Worked examples

A PDF that already has selectable text

You enter
Pdf
1 page with text lines: Meeting notes for the garden club / Next meeting is on Tuesday evening
Language
English
Pages
all
You get
Text
--- Page 1 --- Meeting notes for the garden club Next meeting is on Tuesday evening

The text was taken straight from the file, without OCR. The two lines were joined with a space, so line breaks are lost on pages that use their own text.

Run through the tool’s own code.

A scanned page read with OCR

You enter
Pdf
1 page that is only a picture of the words Invoice 2026 / Total due 45 euros
Language
English
Pages
all
You get
Text
--- Page 1 --- Invoice 2026 Total due 45 euros

No text could be selected, so the page was drawn as an image and read by OCR. This test picture was clean, large type. Real scans are messier, so always check the result.

Run through the tool’s own code.

A file that mixes scans and real text

You enter
Pdf
3 pages: a picture of text, a page with real text, a picture of text
Language
English
Pages
all
You get
Text
--- Page 1 --- Invoice 2026 Total due 45 euros --- Page 2 --- This page has real selectable text inside it --- Page 3 --- Hello scanned world Second line here

Page 2 had its own text, so it was copied directly. Pages 1 and 3 were read by OCR. The result keeps the original page order.

Run through the tool’s own code.

The wrong language loses accents

You enter
Pdf
1 scanned page showing Camión rápido / El niño está aquí
Language a
English
Language b
Español
You get
English
Camion rapido El nino esta aqui
Español
Camión rápido El niño está aquí

With the English model the accents were dropped. The exact wrong letters can change with the picture and the font. With the Spanish model the text came out right. Pick the language that matches the printed page.

Run through the tool’s own code.

Choosing pages with a range

You enter
Pdf
4 pages
Pages
3, 1-2
You get
Pages read
1, 2, 3
Page skipped
4

The numbers are put in order and any repeat is merged. Page 4 is not read. A number outside the file, such as 0-2 or 2-9 for a 4-page file, gives the message Choose page numbers from 1 to 4.

Run through the tool’s own code.

A file with more than 120 pages

You enter
Pdf
121 pages
Pages
1-3
You get
Status
For safety, this tool supports up to 120 pages at a time. Use the page range to process a smaller section.

The page count of the whole file is checked before the range is used. So a 121-page file is refused even when you ask for three pages. A 120-page file with range 119-120 worked.

Run through the tool’s own code.

When it helps

  • Copy the words from a scanned letter, form or printed page into an email or a document.
  • Get the text of a scanned article or book chapter so you can search it or quote from it.
  • Pull the text out of a PDF that does not let you select words with the mouse.
  • Turn a few pages of a scanned contract or report into text you can read on a small screen.
  • Read a scanned handout in one of the 16 languages offered on the page.
  • Read only one chapter of a long scanned book by typing its page range.

Limits to keep in mind

  • OCR languages are English, Spanish, French, German, Portuguese, Arabic, Simplified Chinese, Japanese, Korean, Hindi, Bengali, Russian, Urdu, Turkish, Vietnamese and Indonesian. Only one can be chosen per run, and the page cannot detect the language itself.
  • The whole file may have at most 120 pages, even if you ask for fewer with a page range. For a longer file, split it first with a PDF tool.
  • A page is treated as real text only if it has more than 20 selectable characters. Any other page is drawn as an image and read by OCR, even if a few stray characters are on it.
  • Pages are drawn at up to 1.7 times their size and no wider than about 1,800 pixels before reading, so very small print, faint ink, columns, tables and handwriting are often misread.
  • The result is plain text only. Layout, images, tables and fonts are not kept, text-based pages come out without line breaks, and no searchable PDF is made.
  • Pressing Cancel stops the reading after the page in progress. The text of the pages already read is kept, and you can press Extract again.

Tips

  • Try one or two pages first by typing a range such as 1-2. Check the language and the quality, then run the whole file.
  • Choose the language of the printed text, not the language of your screen. A wrong choice drops accents and misreads letters.
  • Use a straight, sharp, well-lit scan. Skewed, blurry or shadowed pages give worse text.
  • Read the result against the page, especially for numbers, dates, names and amounts, where one wrong digit matters.
  • Split a long file into parts of fewer than 120 pages, and process the parts one after another.
  • Keep this tab open until the scan ends, because long files can take a while on a slow device.

Sources

Common questions

About PDF OCR & Scanned Text Extractor.

Can I copy text from a scanned PDF?

Yes. Choose the file and the language, and press Extract text / OCR. Pages that are only pictures are read by OCR, and the text appears in the box for copying. Recognition can make mistakes, so check the result against the page.

Is my PDF uploaded to a server?

No. The PDF is opened and read in your browser. The OCR engine and the language file are loaded from this same site's own files the first time they are needed. The content of your PDF is not sent anywhere.

Which languages can it read?

It offers 16: English, Spanish, French, German, Portuguese, Arabic, Simplified Chinese, Japanese, Korean, Hindi, Bengali, Russian, Urdu, Turkish, Vietnamese and Indonesian. You pick one for each run, so a page with two languages is best split into separate runs.

Why do I have to choose a language?

The reading engine uses a language file to know which letters and words to expect. A wrong choice gives wrong results, such as lost accents. The language only matters for scanned pages, because real text is copied as it is.

Why was the first run slow?

On the first scanned page the browser has to download the OCR engine, a few megabytes, and the language file, which is roughly 0.4 to 2 MB for the languages offered. The page says your browser keeps the file, so later runs are quicker unless the cache is cleared.

How does the tool decide a page needs OCR?

It first tries to read the text stored in each page. If a page has more than 20 characters of selectable text, that text is used. If not, the page is drawn as an image and read with OCR. If all pages have text, no OCR runs and no language file is loaded.

Does it create a searchable PDF?

No. It gives you plain text to copy, and your original PDF is not changed. If you need a PDF with a hidden text layer, you need a different tool.

Why do some pages come out as one long line?

Pages that already have their own text are read item by item and joined with spaces, so their line breaks are not kept. Pages read with OCR keep the line breaks that the reader found in the picture.

How accurate is the result?

It depends on the scan. Clean, large, straight print in the right language reads well. Small print, faint ink, handwriting, columns and tables are more often wrong. Always proofread names, dates and numbers before you use the text.

Is there a limit on the number of pages or the file size?

The file can have at most 120 pages in total. The page code sets no size limit in megabytes, but big files need more memory on your device and take longer, since pages are read one after another.

How do I save the text to a file?

Press Download TXT to save a plain text file, or press Copy text and paste it into an editor. The Extracted text box is read-only, so you cannot edit inside it. Each page starts with a line like --- Page 1 ---, which you can delete in your editor.

Can it read handwriting or tables?

Not reliably. The reading engine is built for printed text. Handwriting is often misread, and tables come out as lines of plain text without their rows and columns. Use the result as a draft that you correct by hand.

Is my scanned PDF uploaded?

No. OCR runs in your browser. The selected language data is loaded from UsefulVerse on first use and may be downloaded again if your cache is cleared.