Skip to content

OCR PDF

Make the text in scanned pages searchable.

Drop a PDF here

Only your browser reads them.

How to use OCR PDF

  1. Add a scanned PDF. Drop your scanned document onto the upload zone or choose a file from your device.
  2. Select pages to process. Leave Pages blank to scan the whole file, or enter a range, and keep Skip pages that already have text checked for hybrid PDFs.
  3. Run OCR locally. Click Make searchable to load the self-hosted English recogniser and detect words on each scanned page.
  4. Save the searchable PDF. Download the updated file with its invisible text layer ready for Ctrl+F search and copy-paste.

Your files are processed on this device and never uploaded. They stay private, and there's nothing to delete afterwards.

How does an invisible text layer make scanned PDFs searchable?

Running OCR here creates a “sandwich” PDF: your original scanned page image stays on top unchanged, while an invisible layer of selectable text sits directly underneath each word. Every pixel and word box is processed in browser memory with 0 bytes uploaded, letting you search with Ctrl+F and copy passages in any viewer.

Rasterizing at 300 DPI and isolating graphics state streams

Before text recognition begins, each page is rendered onto an offscreen canvas via the MDN HTML5 Canvas API Documentation at roughly 300 DPI, capped at 5,000 px on the longest edge so massive blueprints never exhaust memory.

Once word bounding boxes are detected, the PDF writer wraps existing page graphics in q and Q state operators and draws invisible Helvetica text using PDF text rendering mode 3, as defined in the Library of Congress PDF 1.7 (ISO 32000-1) Format Description.

Why page appearance stays identical in sandwich PDFs

Because the original image stream is never re-compressed or replaced, signatures, stamps, and paper textures look 100 percent identical to the source scan. After making a scanned contract searchable, you can permanently black out sensitive terms by selecting the newly recognized text instead of drawing blind boxes.

Readers can also highlight clauses, copy serial numbers, and index the PDF inside desktop search tools without altering its visual layout.

Running Tesseract WASM offline without uploading sensitive scans

Uploading passports, tax returns, or medical records to cloud OCR sites exposes private documents to external servers. Here, the recogniser downloads ~5 MB of self-hosted WebAssembly and English LSTM data directly from this site on first use, then runs offline in your browser with zero network requests.

Processing stage Local browser action Technical threshold Privacy guarantee
Page inspection Checks existing text items Skips pages with text 0 bytes leave device
Canvas render Scales page for OCR ~300 DPI, max 5,000 px Runs in local RAM
LSTM recognition Runs Tesseract.js WASM Filters confidence > 30 Self-hosted, no CDN
Text layer write Appends mode 3 text Standard Helvetica font Original scan intact

Following the Tesseract OCR Official Documentation, scaling pages near 300 DPI gives the LSTM neural network crisp character strokes and reliable line segmentation. Filtering out noisy artifacts with confidence 30 or below prevents stray speck marks from polluting copy-paste output.

Once the recogniser finishes, the page canvas is cleared immediately from browser memory so large multi-page scans do not accumulate stale bitmaps. If a malformed or truncated PDF scan refuses to open in the viewer, fix files that won’t open first to rebuild its cross-reference table before running OCR.

Skipping existing text pages, filtering ranges, and improving accuracy

Real-world office packets often mix digitally exported cover letters with scanned attachments. Instead of wasting time re-scanning pages that already work, you can target specific pages and optimize scan contrast beforehand.

Using Skip pages that already have text on hybrid documents

Keeping Skip pages that already have text checked inspects each page’s text stream before loading the recogniser. Pages that already contain selectable text are bypassed automatically, while scanned image pages receive the hidden text layer.

If you only need text from pages 4 to 8 of a 50-page packet, enter that range in Pages before clicking Make searchable. Targeting specific page spans avoids waiting on cover sheets or blank separator pages, and lets you process large scanned binders in focused batches without straining browser memory.

Once complete, the output summary reports the exact word count recognized alongside how many pages were skipped for already containing text, giving you a clear verification check before saving.

Straightening crooked scans before running OCR

Tilted feeder scans and skewed phone photos reduce LSTM line-detection accuracy because text baselines cross multiple horizontal rows. Running crooked page correction before archiving straightens skewed scans first so bounding boxes align cleanly with the hidden text layer.

Clean, high-contrast black-and-white scans at 300 DPI consistently yield sharper word boundaries than low-light smartphone photos.

Tip: Wait a few seconds per page on large scans while the local Tesseract worker processes each sheet; cancelling mid-run aborts cleanly without leaving partial files behind.

When should you use OCR PDF instead of converting straight to Word?

Choosing between adding a searchable PDF text layer and converting to an editable document depends on your goal: archival visual fidelity versus heavy text editing. Both workflows serve different stages of document handling and complement each other cleanly.

Preserving archival layout vs extracting editable paragraphs

Under the Library of Congress PDF/A-2 Archival Format Description, long-term archives preserve exact visual page scans paired with searchable Unicode text layers. Use OCR PDF when legal, accounting, or compliance records must keep their exact original stamps, handwritten signatures, and letterhead layout while supporting full-text search.

Because the hidden text layer sits directly behind each scanned word, readers can highlight clauses, copy invoice numbers, and verify totals without risking reflowed tables or shifted page breaks.

Exporting recognized scans into editable documents

When you need to rewrite clauses, reformat tables, or edit paragraphs inside a word processor after recognizing a scan, run OCR PDF first and then convert the searchable result with PDF to Word. Running OCR beforehand ensures the Word converter receives clean text coordinates rather than flat bitmap pages.

Standard WinAnsi Latin characters map cleanly into the hidden Helvetica layer, ensuring searchable English words copy accurately across every major PDF reader.

Questions

Is my scanned PDF sent to a server during OCR?

No. The Tesseract.js recogniser downloads once from this site with zero third-party CDN requests and runs entirely inside your browser tab, so it even works offline afterwards.

Does OCR change how my scanned pages look?

No. Your original scanned page images stay visually identical while an invisible PDF text layer is aligned underneath each recognized word for searching and copying.

Which languages are supported by OCR PDF?

English is currently active in the Language selector using the self-hosted Tesseract LSTM English traineddata model. Support for additional Western European and multilingual OCR language packs is planned for future updates.

How accurate is browser OCR on printed vs handwritten text?

Accuracy is high on clean, high-contrast printed documents such as contracts, invoices, and books. Cursive handwriting, very small print, and blurry phone photos are harder to recognize reliably.

Why does the first OCR run download about 5 MB?

The first run fetches the self-hosted Tesseract.js WebAssembly core and English LSTM language data (about 5 MB total) from this site so your browser can run OCR locally without uploading documents.

What happens if a PDF already contains selectable text?

When Skip pages that already have text is checked, the tool inspects each page first and skips pages that already carry real text, saving time on hybrid PDFs.

Updated