How to OCR a PDF

How to OCR a PDF, step by step

A scanned PDF is a stack of photographs: you cannot search it, copy from it or select a single line. OCR reads what is in those images and gives the text back. On the BestPDFInTheWorld OCR tool it takes three steps, with nothing to install:

1. Choose the file. Drag the scanned PDF into the tool window, or an image of a document (PNG, JPEG or WebP).
2. Choose the language of the document. Portuguese, English or Spanish. It is the language of the document, not of the interface, and choosing wrong degrades the result badly.
3. Download the result. You get a searchable PDF with the original pages, or the plain text in a .txt file.

The whole job happens on your computer, start to finish: the scan never leaves the browser.

What OCR gives you back

By default you get a searchable PDF: the original document, pages untouched, plus an invisible text layer on top that the reader does not draw. Nothing new appears on the page, but your PDF reader's search finds the words, the mouse selects lines and copy and paste works. It is also how search engines and internal archives start finding the document.

If you prefer the text alone, the tool gives it back as plain text in a .txt file. An honest note: recognition reads lines of text, not the structure of the page, with a two column document, or a dense table, the order often comes out scrambled in the .txt. That does not show in the searchable PDF, because each word stays where it belongs.

When to use OCR

• Use it when you cannot select anything. If your PDF reader cannot mark a single word, the document is made of images and OCR is the right tool.
• Skip it if the PDF already has text. If you can select a word, the text is already there: use the Extract text tool instead, which is instant and invents nothing.
• It does not read handwriting. A signature, a handwritten note in the margin or a form filled in by hand are not recognised, whatever comes out of them is noise.
• It only reads the three languages it has a model for. Portuguese, English and Spanish. A document in French or German will be read with whichever model you pick, and the result will be poor. A document mixing two languages is read as the one you pick, and the other comes out worse.
• The result is only as good as the scan. A straight page, decently lit and at a decent resolution, reads back almost perfectly; a photograph taken at an angle, with a shadow across it, or scanned at low resolution, gives text with errors, sometimes plenty. If you can scan it again, do: it pays off more than any option in this tool.

Tips for getting the most out of OCR

• Pick the language of the document carefully. It decides the recognition model; the interface language makes no difference to the result.
• The first run takes longer. The first time around, your browser downloads the recognition engine and the language model you picked, which together are several megabytes. It is then kept in the browser: next time it starts straight away. Switch language and it downloads that language's model, and only that one.
• Expect a few seconds per page. It depends on the machine and the quality you pick. The tool reports progress page by page so you can see it moving. Long documents genuinely take a while: that is the price of reading letter by letter.
• Start with Fast quality. It is for checking quickly whether a document is worth processing. Normal covers most cases; High helps with small type or weak scans, and takes considerably longer.
• The text layer is an approximation. Words land on top of the words in the image, but the selection box can sit slightly off, especially with very small type or on skewed pages. That does not affect search: it affects the rectangle your reader highlights when it finds the word.
• Scanned on your phone? The pages that come out of the Scan to PDF tool are images, exactly as they left the camera: run the result through OCR to make it searchable.

This tool is part of the Premium subscription, EUR 1.99 a month, which removes the limits on every tool and adds OCR. Sign PDF opens on the free plan, with one signature a day.

Sources

• BestPDFInTheWorld, OCR de PDF digitalizado
• Tesseract.js
• MDN, File API