How to Translate a Scanned PDF with OCR

The Problem with Scanned PDFs

When you scan a paper document and save it as a PDF, the file contains only an image of the page — not selectable, copyable text. Regular translation tools fail completely because there is nothing to translate: the text exists only as pixels.

This affects a huge range of documents: notarized contracts, government-issued certificates, historical records, fax printouts, and any document that arrived as a physical paper and was then digitized. In 2026, scanned documents still represent a significant share of the files businesses need to translate.

How OCR + AI Translation Works

The solution is a two-step pipeline: Optical Character Recognition (OCR) followed by neural machine translation.

1. OCR analysis — the engine scans each page pixel by pixel, identifies character shapes, and reconstructs the text layer. Modern OCR handles 100+ scripts including Latin, Arabic, Cyrillic, CJK (Chinese, Japanese, Korean), Devanagari, and more.
2. Layout detection — columns, tables, headers, and footers are identified and mapped so that the translated output can be placed back into the same zones.
3. Text extraction — the reconstructed text is cleaned: hyphenation artefacts, line-break noise, and scanning imperfections are corrected before translation.
4. AI translation — the clean text is passed to a neural translation model fine-tuned for document context, preserving paragraph and sentence boundaries.
5. Format reconstruction — the translated text is laid out in the original document structure and exported as a PDF.

Image Quality Requirements

OCR accuracy depends directly on scan quality. These conditions produce the best results:

• Resolution — 300 DPI or higher. 150 DPI is passable for large print; below 100 DPI accuracy drops sharply.
• Orientation — pages should be straight. Auto-deskew corrects up to ~10° of tilt but larger skews degrade results.
• Contrast — dark text on a light background. Faded ink, coloured paper backgrounds, or watermarks reduce accuracy.
• No handwriting — printed text is recognized with high accuracy; handwritten annotations are extracted where legible but flagged as uncertain.
• File format — PDF, JPG, PNG, and TIFF are all accepted. Multi-page PDFs are processed page by page.

Step-by-Step: Translate Your Scanned PDF

1. Upload the scanned PDF — drag and drop or paste a file URL. The system detects automatically that it is a scanned (image-based) document.
2. Select source language (optional) — auto-detect works for most common scripts. Set it manually for rare languages or mixed-script documents.
3. Select target language — 100+ languages available.
4. OCR runs first — text is extracted and a preview of the recognized content is shown. Spot-check for obvious errors before proceeding.
5. AI translation runs — the extracted text is translated with full document context preserved.
6. Download translated PDF — the output file has the same page dimensions and column layout as the original scan.

Common Scanned Document Types

• Notarized contracts — signed and stamped documents that cannot be re-typed without losing authenticity
• Government-issued certificates — birth, marriage, academic, and professional certificates from official bodies
• Court records and legal filings — historical case files, archived judgments
• Medical records — patient files, lab reports, referral letters from older healthcare systems
• Financial statements — legacy bank statements, audit reports printed before digital archiving
• Technical manuals — product documentation from manufacturers that shipped only print copies