OCR PDF: make scans searchable

When does a PDF need OCR?

A PDF needs OCR when you can't select or search its text, which means the pages are only pictures of paper. You can read such a scan, but your computer can't: Find returns nothing, you can't copy a paragraph into an email, screen readers stay silent, and document management systems can't index it. OCR fixes that for contracts, invoices, old letters, book chapters, receipts and anything that came out of a scanner or phone scan app.

If you can already select text in a PDF, it doesn't need OCR. That's why any page that already holds 50 or more text characters is skipped automatically, which keeps mixed documents fast and avoids a duplicate text layer. Tick the option to OCR those pages anyway when their existing text is garbled, for example after a poor earlier OCR run.

How does OCR PDF work?

OCR PDF renders each page as an image at about 250 DPI with pdf.js and passes it to Tesseract, an open-source OCR engine running as WebAssembly in your browser. Tesseract's neural-network recognizer returns every word with its position, baseline and a confidence score, and words scoring below 15 out of 100 are discarded as noise.

Those positions are mapped back onto the original PDF page, taking page rotation and cropping into account, and the words are written as invisible text exactly over the printed ones. Because the original page content is kept rather than replaced with a new image, the scan's quality and file structure stay as they were, and the file grows only by the size of the text, typically a few kilobytes per page.

Should I make a searchable PDF or convert to Word?

Make a searchable PDF when the document must keep its exact appearance, and convert to Word only when you need to rewrite the content. A searchable PDF keeps the scan exactly as it looks, which matters for signed contracts, forms, certificates and anything you may need to prove later. It is also the format most archives, courts and document systems expect.

In a Word conversion the layout is rebuilt from the recognized text, so tables, columns and fonts rarely match the original perfectly. A common workflow is to OCR the scan here, keep that PDF as the record, and send a copy through PDF to Word for editing. When only the raw wording is needed, PDF to Text extracts the new text layer into a plain file.

How do I get better recognition?

Better recognition comes mostly from a better scan, because the engine can only read what is clearly visible on the page. Sharp, straight, evenly lit pages with dark text on a light background are read almost without errors, while shadows, curved book pages and heavy JPEG artefacts cause misread letters. These steps help the most:

• Scan at 300 DPI in greyscale or black and white. Lower resolutions lose small print.
• Keep pages straight and evenly lit. Phone scan apps that flatten and crop the page work much better than plain photos.
• Rotate sideways pages with Rotate PDF before running OCR.
• Large photo scans can be shrunk with Compress PDF first; OCR works on the page as it looks, so a moderate recompression rarely hurts accuracy.

What are the limits of OCR PDF?

OCR PDF reads printed English only, and it does not recognize handwriting reliably. Complex layouts such as multi-column tables may copy out in an unexpected order even when every word is found. The invisible layer uses the standard Helvetica font, so ligatures and unusual dashes are simplified to plain letters and hyphens, and characters outside the Western character set are stored as question marks.

Files up to 100 MB are accepted, and very long scans are best split into parts on phones, where memory is tighter. If no words are found at all, the scan is probably blank, too faint, below about 150 DPI, or not in English. Recognition can be cancelled at any point, and nothing is saved until the final page has been read.