What OCR does to a PDF
OCR, optical character recognition, turns a picture of letters into real text. This tool runs Tesseract, an open source OCR engine, in your browser through tesseract.js. It draws each page of your PDF as an image, finds the lines and words on it, and reads them. Then it writes the words back into the PDF, each one at the spot where it appears on the page, but invisible. You still see the original scan, and on top of it lies text that you can search with Ctrl + F, select and copy.
Why a scanned PDF is not searchable
A scanner takes a photo of the paper. The PDF it saves holds that photo, not the letters. As the US Bankruptcy Court for the District of Nevada puts it in its filing instructions, scanned documents are essentially image files and do not contain recognizable text by default. The court's quick test works in most PDF readers: press Ctrl + A. If single lines of text are highlighted, the PDF has text. If the whole page is highlighted as one block, it is a picture. That is the kind of page this tool is for. If your PDF already has text, PDF to Text extracts it directly without OCR.
How the searchable PDF is built
The tool does not redraw your pages. It opens your PDF, keeps every page as it is, and adds the recognized words to each page in the invisible text mode of the PDF format, text rendering mode 3, which neither fills nor outlines the letters. Tesseract's own PDF output uses the same mode for its text layer. Each word is stretched to the width it has in the scan, so a selection lines up with the printed word, and a space follows every word so that copied text keeps its spaces.
The text uses Helvetica, one of the standard fonts every PDF reader has built in, so no font is embedded. That is why the file grows only a little. The text layer can hold the letters of the six supported languages; a rare character outside that set is written as a question mark.
The settings
- Languages: pick every language in the document. The engine expects the letters and words of the languages you choose, so the right choice matters for accents and umlauts.
- Resolution: each page is drawn at 200 or 300 dpi before reading. The Tesseract documentation says it works best on images of at least 300 dpi. 200 dpi is faster and needs less memory, which helps on phones and with large print. Very large pages are drawn with fewer pixels if your browser cannot hold them.
- Pages to read: leave the field empty for all pages, or type pages and ranges like 1-5, 8.
- Skip pages that already have text: before reading, the tool asks every page for its text. A page with at least 40 characters other than spaces counts as having text and is skipped when this is ticked.
Tips for better accuracy
- Turn pages upright first. Pages are read as they are shown. A sideways page gives nonsense, so fix it with Rotate PDF. The Tesseract documentation notes that line detection gets significantly worse when a page is too skewed, so rescan crooked pages if you can.
- Watch for dark borders. The same documentation warns that the dark borders around scanned pages can be picked up as extra characters.
- Tables are hard. The documentation says Tesseract has known problems reading text and data from tables. Check figures in tables by hand.
- Read before you compress. Run OCR on the full-quality scan, then make the result smaller with Compress PDF. The guide how to fix a scanned PDF explains the best order of steps.
Privacy: what is downloaded and from where
Your PDF stays on your device. Drawing the pages, reading the text and saving the new file all happen in your browser. Two kinds of files are downloaded from this website, freebiefy.com, not from a third-party server: the OCR engine (tesseract.js with its WebAssembly core, about 3.8 MB) and the language data from freebiefy.com/models/tesseract/, between 0.7 and 2.8 MB per language. They are the same files the Image to Text tool uses, and the tool shows their size before the first download. Nothing about your file is sent back.
Limits
- Encrypted PDFs with a password or editing restrictions cannot be changed. Save a copy without protection first.
- Long documents take time, because every page is read one after another. You can read a range of pages at a time.
- Memory on iPhone and iPad is limited. If a large document stops with an error, try 200 dpi or fewer pages at once.
How to make a scanned PDF searchable
- 1Drop your scanned PDF on the box. The tool counts the pages and checks which ones already have text.
- 2Choose the languages, the resolution and, if you like, a page range.
- 3Press Make PDF searchable. The first time, the OCR engine and language data download. Then the pages are read one by one with a progress bar.
- 4Download the searchable PDF, or only the text as a .txt file. Open the PDF and press Ctrl + F to search it.
Sources
- Tesseract documentation, Improving the quality of the output: works best at 300 dpi or more, skewed pages reduce line segmentation, dark scan borders read as extra characters, problems with tables
- Tesseract source code, pdfrenderer.cpp: Tesseract's own PDF output writes its text in rendering mode 3 ("3 Tr", invisible ink), stretched to each word, with a space between words
- US Bankruptcy Court, District of Nevada: Text searchable PDFs: scanned documents are image files without recognizable text, and the Ctrl + A check
Related guides
- How to Fix a Scanned PDF: The Right Order of Steps
- Copyright in the Classroom: What Teachers May Copy
- How to Scan Documents With Your Phone Into a Clean PDF
Frequently asked questions
What does OCR PDF do?
It reads the text on every scanned page with the open source Tesseract engine and writes that text into the PDF as an invisible layer on top of the page. The page looks exactly as before, but you can now search it, select words and copy them.
Is my PDF uploaded?
No. Your browser draws the pages, reads the text and saves the new PDF on your device. The OCR engine and the language data are files from this website that your browser downloads the first time; your PDF never leaves your device.
Why is the new PDF larger than the original?
Because the text layer is added to the pages. The pages themselves are kept as they are, not redrawn or recompressed, so the scans keep their quality and the file grows only by the text and its positions, usually a few KB per page.
How accurate is the text?
It depends mostly on the scan. Sharp, straight pages scanned at 300 dpi or more give good results. Crooked pages, dark borders, tables, tiny print and decorative fonts cause errors. The tool shows the confidence the engine reported. Check important numbers and names in the text before you rely on them.
What about pages that already have text?
The tool checks every page first. Pages with at least 40 letters, digits or signs of real text are counted as having text, and you can skip them. If you read them anyway, they get a second text layer and a search finds those words twice.
Which languages are supported?
English, German, French, Spanish, Italian and Portuguese. Choose every language that appears in the document. Each language is a separate download of 0.7 to 2.8 MB.