PDF OCR

PDF OCR recognizes printed text on rendered page images. It is meant for scans and image-only pages. Inspection peeks at page 1’s text layer and may note that selectable text already exists.

• You need a working copy of scan text to proofread.
• PDF to Word or Excel reported no extractable text.

• Recognize printed English (or another listed language) on scanned pages.
• Limit OCR to first, first 3, or a custom parsePageRanges list on a long file.
• Export Markdown with # Page N headings when you want labeled sections.

Pages are rendered at scale 1.4 (Fast), 2 (Standard), or 2.5 (Accurate). Standard and Accurate enable a contrast/grayscale canvas pass unless you disable enhance. createWorker(language) runs recognize() per page. Languages include eng, deu, fra, spa, and several others listed in OCR_LANGUAGES.

Paged

--- Page N --- separators in a .txt file.

Plain

Pages joined with blank lines, .txt.

Markdown

# Page N headings, .md.

Input

One PDF, up to 50 MB.

Pages

All, first, first 3, or a custom parsePageRanges list.

Output

{base}-ocr.txt or {base}-ocr.md

1. Upload the scan.
2. Match language to the print language.
3. Pick Fast, Standard, or Accurate.
4. Pick page scope and Paged, Plain, or Markdown.
5. Click Run OCR.
6. Proofread the download against the PDF.

You get recognized text, not a layout-faithful document. Handwriting, low-contrast scans, skewed pages, and mixed languages can fail. The tool does not claim legal or archival accuracy.

pdf.js and Tesseract.js run in the browser. Tesseract may still fetch language/worker assets as the library implements. The PDF is not posted to a DevToolKit OCR API.

• PDF to Word/Excel require a text layer and do not run Tesseract.
• PDF to Image exports page pictures without recognition.

• Compare every page to the scan.
• Treat average confidence as a hint only.
• Re-run with Accurate on faint type if needed.
• Remember the download is {base}-ocr.txt or {base}-ocr.md, not a new searchable PDF.