How to OCR Scanned PDF to Text Offline

About This PDF Conversion Tutorial

This guide was written and tested by David Kim, a CAD & Engineering Documentation Lead with 10 years of hands-on experience in CAD conversion, engineering drawings, technical documentation. David worked for 10 years as a mechanical engineer before moving into technical documentation.

Scanned PDFs are the digital equivalent of a photocopy — a picture of text that cannot be searched, selected, or edited. OCR (Optical Character Recognition) converts these frozen images into live text that can be searched, copied, and edited. The difference between a good OCR result and a poor one is not just about accuracy — a 95% accurate OCR means 5 errors per 100 characters, or roughly one error per sentence.

This guide compares four methods for OCR on scanned PDFs. We tested each on typed documents, handwritten notes, low-quality photocopies, and mixed-language documents to measure recognition accuracy and layout preservation.

Time to read: 10-12 minutes | Difficulty: Intermediate | Last updated: 2026-07-27

Why Conversion Fidelity Decides Whether the File Is Usable

OCR engines work by analyzing pixel patterns in an image and matching them to known character shapes. Modern OCR engines use neural networks trained on millions of documents to recognize text in various fonts, sizes, and languages. The quality of the input scan is the biggest factor in OCR accuracy: 300 DPI minimum is recommended, and clean, high-contrast scans produce much better results than wrinkled or faded pages.

OCR has two output modes: text layer (the recognized text is stored invisibly behind the scanned image, making the PDF searchable) and editable text (the recognized text and reconstructed layout replace the scanned image entirely). Searchable PDF mode preserves the original appearance while adding text functionality; editable mode allows full text editing but may alter the visual layout.

Converting with a Desktop PDF Editor, Step by Step

📋 Quick Recap of the Conversion Steps

Print this box or keep it open while you work through the tutorial.

1. 1Launch your desktop PDF editor and open the scanned document. If the editor has image preprocessing, run it: de-skew (straighten t…
2. 2Choose the language(s) containing all text in the document. If the document has passages in multiple languages, select all applica…
3. 3Click Run or Recognize. Processing time depends on page count and DPI — a 50-page 300 DPI document may take 1-2 minutes. After com…
4. 4If the same error appears throughout the document (e.g., 'accounts' always recognized as 'acc0unts'), use the OCR correction tool …
5. 5Save the OCR-processed file with 'Searchable' or 'OCR' in the filename to distinguish it from the original. Open in a different PD…

Free Conversion Options, With Honest Limitations

No desktop editor installed? Here are free alternatives — we list their limitations honestly.

Method 1: Tesseract OCR (Free, Command-Line)

Open-source OCR engine. Run: tesseract scanned.pdf output -l eng. Extracts plain text. Limitation: command-line only; no layout preservation; produces plain text, not searchable PDF; lower accuracy on complex layouts.

Method 2: Google Drive (Free, Requires Upload)

Upload PDF to Google Drive → Right-click → Open with Google Docs. Google's OCR engine processes the file. Limitation: file is processed on Google servers; limited to 10 pages in free version; output is a simplified document without the original layout.

Method 3: Microsoft OneNote (Free, Windows/Mac)

Insert PDF as printout → Right-click image → Copy Text from Picture. Limitation: text only — no layout preservation; one page at a time; accuracy is moderate.

OCR Accuracy by Document Quality

Language-Specific OCR Tips

English: Most OCR engines are optimized for English — expect the highest accuracy rates. Enable the English dictionary for automatic spelling correction.

Multi-language documents: If a document contains English with French, German, or Spanish passages, enable multi-language OCR. The engine will detect and switch languages per paragraph. Accuracy drops slightly (~2-3%) compared to single-language OCR.

CJK (Chinese/Japanese/Korean): Requires a CJK-capable OCR engine. Accuracy is lower than Latin scripts (~90-95% for clean 300 DPI scans) because CJK characters have more complex stroke patterns and there are thousands of distinct characters to distinguish. Enable the correct language pack before running OCR.

Right-to-left scripts (Arabic, Hebrew): Enable RTL text direction detection. OCR accuracy is generally good (~95%) for clean printed text, but handwritten Arabic has among the lowest OCR accuracy rates (~70-80%) due to connected letterforms and diacritical marks.

Post-OCR Quality Assurance

1. Run spell check on the OCR output. Most desktop editors include a built-in spell checker that highlights suspicious words — work through these systematically.
2. Check numbers and dates carefully. OCR engines frequently confuse: 0/O, 1/l/I, 5/S, 8/B. For financial or legal documents, manually verify all critical numbers against the original scan.
3. Look for "tofu" (□ boxes). These indicate characters the OCR engine couldn't recognize. They often appear for special symbols (©, ®, ™, math symbols) or characters from scripts not enabled in the OCR language settings.
4. Compare paragraph counts. If the original scan has 12 paragraphs and the OCR output has 11, a paragraph was likely merged — search for unusually long paragraphs and separate them at logical break points.
5. Check text ordering on multi-column pages. OCR engines sometimes read across columns (line 1 column A → line 1 column B → line 2 column A) instead of down columns. Enable "column detection" or "zone OCR" to prevent this.

In-Depth: Scenarios & Best Practices

OCR Accuracy: What to Expect

Clean printed text (laser-printed, 300+ DPI scan): 99%+ accuracy. Modern OCR engines handle standard fonts nearly flawlessly on high-quality scans. Remaining errors are usually punctuation or special characters.

Typewriter documents (older, lower contrast): 95-98% accuracy. Uneven ink density and monospaced fonts challenge OCR engines. Expect scattered errors on worn or faded characters. Manual proofreading is essential.

Handwritten text: 70-85% accuracy on neat handwriting, below 60% on cursive. Budget time for significant manual correction.

Mixed content (text + tables + images): Text accuracy is high (98%+), but table structure recognition varies. Simple tables with clear gridlines convert reliably; complex tables with merged cells or nested headers often require manual reconstruction.

Post-OCR Cleanup Checklist

Every OCR output needs review: (1) Run spell-check on the entire document. (2) Search for common OCR confusion pairs: "rn" vs "m", "cl" vs "d", "0" vs "O". (3) Verify all numbers — dates, dollar amounts, phone numbers. (4) Check paragraph breaks — OCR often inserts extra line breaks. (5) Verify heading styles — OCR output typically has all text at the same level.

Desktop vs. Online Conversion: Which Preserves More?

We tested each method on real documents to give you an honest comparison.

Verdict: Online tools are fast for simple conversions. For complex documents with tables and images, desktop editors preserve the most formatting.

Why Desktop OCR Engines Produce Better Results

Desktop PDF editors embed professional OCR engines that use neural network models trained on large document datasets. They support multiple languages (with automatic language detection), layout preservation (columns, tables, and images), and image preprocessing (de-skew, de-noise, contrast enhancement) that improves recognition on poor-quality scans. OCR processing happens locally, so there are no file-size limits and no privacy concerns.

Online OCR services typically use older OCR engines with lower accuracy on complex layouts. They also impose page limits (often 5-10 pages for free tiers) and file-size caps, and require uploading potentially sensitive documents to external servers. See the full comparison →