Tesseract
The most widely used open-source OCR (optical character recognition) engine — originally developed at HP, now maintained by Google.
Tesseract is the de-facto open-source OCR engine. Originally developed at Hewlett-Packard between 1985 and 1995, open-sourced in 2005, and now maintained by Google, it powers OCR in countless applications — from Evernote's scanning to academic digitization projects to our own OCR PDF tool.
What Tesseract does
Given an image of text — a scan, a photograph, a screenshot — Tesseract produces:
- The recognized text as a string
- Bounding boxes for each character/word/line
- Confidence scores per recognition
- Optionally, a "searchable PDF" overlay — the original image with an invisible text layer on top
Languages
Tesseract ships with trained data files for 100+ languages. Accuracy varies significantly:
- Excellent — modern Latin-script printed text (English, most European languages), Arabic, Chinese (Simplified and Traditional), Korean, Japanese
- Good — Cyrillic, Hebrew, Thai, Vietnamese
- Workable — handwritten English (limited), historical scripts
- Challenging — cursive handwriting, heavily stylized fonts, low-resolution scans
Tesseract's strengths and limits
Strong at — clean, printed, well-lit scans at 300 DPI or higher. Consistently >98% character accuracy on standard office documents.
Limits — handwriting (consider Google Vision API or Azure Document Intelligence instead), tables and layout reconstruction (ABBYY FineReader does this far better), and real-time OCR on mobile (Apple's Vision framework and Google ML Kit are faster on-device).
For most PDF scanning needs, Tesseract is the right tool at the right price (free). That's why it powers our OCR PDF tool.