OCR PDF — Recognize Text in Scans

How to use it

1. 1Choose a scanned PDF.
2. 2Wait while each page is rasterized and recognized — progress is shown per page.
3. 3Read the per-page result and confidence, then copy or download the text.

Limits worth knowing

• Up to 20 pages per run, to keep the browser responsive.
• English (eng) only, via the tessdata_fast model served from this origin.
• OCR quality depends on the scan: clean, upright, ~300 DPI input works best.
• Password-protected PDFs are refused — remove the password first.
• Handwriting and stylized display faces are outside what OCR reads reliably.
• The output is text, not a “searchable PDF” — no text layer is injected into the file.

What OCR actually does here

Each page of the PDF is rasterized at roughly 144 DPI, and the picture is handed to Tesseract.js — the same open OCR engine used on the command line — running entirely in your browser. The result is the text the engine saw, one block per page, with the confidence it reports for that page.

The PDF and recognized text stay in your browser. The OCR worker, WebAssembly engine and English language model load from this site on first use and are cached by your browser.

Reading the confidence number

Tesseract reports its own confidence per page on a 0–100 scale. Above 85% is usually trustworthy for typed English. Between 60% and 85%, expect mistakes — particularly with numbers, punctuation and unusual fonts. Below 60%, proofread everything or get a better scan.

If a page comes out empty or garbage, the page is almost certainly not a clean upright scan — rotate it, straighten it, and re-scan at a higher resolution before concluding the tool is broken.