About OCR PDF
A scanned PDF is a photograph of a document. It looks like text but contains none, which is why you cannot search it, select a sentence, or have a screen reader read it. Optical character recognition, usually shortened to OCR, reads the letters off the picture and writes them back into the file as real text.
This tool does that to a PDF in your browser. It reads each page, finds the words, and places them invisibly underneath the picture, so the page looks exactly the same but can now be searched and copied from. Only the language model used for recognition is downloaded, the first time; your document is not uploaded anywhere.
How to make a scanned PDF searchable
- Choose the scanned PDF.
- Pick the language the document is written in. English is the default, and 85 languages are available, including several non-Latin scripts. Choose the one that matches the document, not your own: recognition is far less accurate in the wrong language.
- Press the button. The first time you use it, the recognition data for that language is downloaded, which can take a few seconds. After that, each page is read in turn, and progress is shown.
- Read the result line. It reports how many words were read and on how many pages, and how many pages were left alone because they already had text.
- Check the preview, then save. Open the saved file and try searching for a word you can see on the page.
When you would use it
- Making a scanned contract, letter or record searchable, so that a word can be found in seconds instead of by paging through.
- Copying a paragraph out of a scanned book chapter or a paper archive.
- Making scans readable by screen readers and other assistive tools.
- Preparing an archive so that a document-management system can index its content.
- Turning a photo-of-a-page PDF into something you can work with afterwards, for example before you redact names from it.
What it will not do
- Recognition is only as good as the scan. A crisp, straight page at 300 dots per inch reads well; a blurred, skewed or low-contrast photograph reads badly, and the result will contain errors.
- It reads printed text. Handwriting is generally not recognised reliably.
- Pages that already contain real text are skipped, and the result tells you how many. That avoids doubling up the words on a mixed document.
- The words are added invisibly under the picture; the visible page is not changed, and mistakes in recognition will not show but will affect search.
- It does not change the file’s size meaningfully. To shrink a large scan, use Compress PDF.
- It works on a PDF, so a photo has to be put into one first with JPG to PDF.
How it compares
Many scanners and phone apps offer OCR as they scan, which is the easiest route if you are still at the scanner. This is for the file you already have, from an inbox or a shared drive, where the scanning is long past.
Cloud OCR services can be very accurate, but they work by sending your pages to a server. Here the recognition runs on your own computer, so a confidential scan stays confidential. The trade is speed: a long document on a modest laptop takes a while, because your machine is doing the work.
Tips
- If the first result is poor, try the document at a better scan quality. Straightening and removing shadows helps more than anything else.
- Do a language check. A French contract read as English comes out as nonsense with a few correct words.
- For a very long document, work in sections. Split it with Split PDF, recognise the pieces, and merge them again.