What OCR does
A scanned page is a picture. You can read it, but a computer sees only dots: you cannot search it, copy a sentence from it, or let a screen reader read it aloud. OCR (optical character recognition) finds the letters in the picture and turns them into real text.
With Searchable PDF, the tool keeps the page images exactly as they are and adds an invisible text layer underneath them. The document looks the same, but now you can search it with Ctrl+F, select and copy text, and find it with the search of your computer or document system. With Plain text (.txt), you get only the recognized text, ready to paste into an email or a document.
Languages
Recognition works much better when it knows the language. Choose every language that appears in the document, up to four. Fewer languages make recognition faster and more accurate. The available languages are:
English, Spanish, Portuguese, French, German, Italian, Dutch, Swedish, Danish, Polish, Czech, Slovak, Hungarian, Romanian, Greek, Russian, Turkish, Indonesian, Vietnamese, Thai, Chinese (Simplified), Chinese (Traditional), Japanese and Korean.
Options for better results
- Straighten tilted pages (on by default): pages scanned at a slight angle are turned straight. This improves recognition and makes the document look neater.
- Remove scanner noise: cleans specks and dirt from poor scans before recognition.
- Pages that already have text: Keep them as they are recognizes only pages without text, which is right for most scans. Redo OCR replaces an old or bad text layer. Rasterize and OCR everything turns every page into an image and recognizes it again; use it when the existing text is garbled.
How to get good OCR
The quality of the scan matters most. Scan at 300 DPI; 200 DPI is the minimum for normal text, and small print needs more. Black and white or grayscale is enough for text. Photos from a phone work if the page is flat, sharp and evenly lit, taken from straight above. Printed text is recognized very reliably; handwriting is usually not recognized. Tables keep their content, but complex layouts with many columns may come out in a different reading order in the text file.
Server processing, limits and privacy
OCR needs a lot of computing power, so it runs on our own server with OCRmyPDF and Tesseract, two well-known open-source tools. Files are sent over an encrypted HTTPS connection, processed and deleted right after; results are removed within 15 minutes. A document can have up to 300 pages and 100 MB. Large scans take a while; the page shows the progress and your place in the queue.
Other tools on this site use OCR too: PDF to Word and PDF to Excel recognize scanned pages automatically, and DjVu to PDF can make the result searchable. To shrink a large scanned PDF after OCR, use Compress PDF.