How to make a scanned PDF searchable with OCR

Do you actually need OCR?

Open the PDF and try to select a single sentence with your cursor.

• If individual words highlight, the PDF already has text. You don’t need OCR; Extract text and Search PDF will work as it is.
• If your cursor draws a box over the whole page, or nothing highlights, the page is an image and you need OCR.
• If it’s mixed, like a typed report with a few scanned signature pages, consider running OCR on just the scanned pages. OCR PDF re-renders every page it processes, including pages that already had perfect text, and replaces that text with its own reading.

Choose the right language

OCR uses knowledge of the language to decide between look-alike characters: “rn” or “m”, “0” or “O”, “l” or “1”. Run a German letter through the English model and the accented letters and common words come out wrong. Always pick the language the document is written in.

OCR PDF currently supports English, Hindi, Spanish, German and French. Each language’s data downloads the first time you use it and is then kept by your browser.

Get a better result from the scan itself

Most OCR errors are decided before you press the button. If you’re scanning the paper yourself:

• Resolution: 300 DPI is a good default for ordinary printed text. Below about 200 DPI, small characters start to merge.
• Straight pages: even a slight tilt increases errors. Use your scanner’s deskew option, or rescan crooked pages.
• Contrast: clean black text on white paper works best. Faint photocopies and coloured backgrounds cost accuracy.
• Phone photos: shoot from directly above, in good light, with the page filling the frame. Scan to PDF builds a multi-page PDF from your photos.

Check what OCR recognised

Even on a clean scan, OCR makes mistakes, and they cluster where they matter most: numbers, names, punctuation, and anything in a table. W3C’s guidance is to verify that the converted text is complete and in the right reading order, not to assume it.

• Search test: press Ctrl+F (Cmd+F on a Mac) and search for three or four words you can see on different pages, including one from a footnote or a table.
• Numbers: search for an amount, date or reference number exactly as printed. This is where 1 and l, 0 and O, or 5 and S get swapped.
• Copy test: copy a paragraph and paste it into a text editor. Check the words come out in reading order, especially on pages with columns.
• Full check: for anything you’ll quote or rely on, compare the output of Extract text with the page itself.

What changes in the file

OCR PDF re-renders every page as a high-resolution image (about 216 DPI) and places each recognised word, invisibly, over the spot where it appears. At normal zoom the page looks the same, but it’s now a fresh image of the original rather than the original page itself, and the file is often larger than before.

Keep your original scan. If you need an editable document rather than a searchable one, converting a scanned PDF to Word builds on this same OCR step.

Limits worth knowing

• Handwriting isn’t reliably recognised. The engine is trained on printed text.
• Supported languages are English, Hindi, Spanish, German and French. For Hindi, use the plain text output.
• The searchable layer drops accents, so search without them.
• Every page is re-rendered, including pages that already had real text, and their text is replaced by the OCR reading.
• OCR runs on your device: slower than a server, and much slower on a phone.
• Password-protected PDFs have to be unlocked first.

Sources

1. W3C Web Accessibility Initiative, PDF7: Performing OCR on a scanned PDF document to provide actual text