Table of Contents
What Is OCR?
How it works
OCR (Optical Character Recognition) is the process of converting images of text into machine-readable text. A scanned PDF is a stack of pictures — pixels. OCR analyzes those pixels, detects character shapes, and maps them to Unicode text. The result: a text layer you can search, copy, and (in some tools) edit.
Modern OCR engines (Tesseract, Google Cloud Vision, Amazon Textract, Adobe's engine) use neural networks trained on millions of font variations. They handle multiple languages, mixed scripts, and degraded images surprisingly well — but they're not magic.
Accuracy factors
- Resolution: 300 DPI minimum. 150 DPI = ~80% accuracy. 72 DPI = garbage.
- Contrast: Black text on clean white background. Gray text, colored backgrounds, bleed-through from backside — all tank accuracy.
- Font: Standard fonts (Times, Helvetica, Arial, Courier) → 99%+. Decorative, handwriting, cursive → 50–80%.
- Layout: Single column, clear margins → great. Multi-column, tables, text wrapped around images → needs layout analysis.
- Language: Engine must support the language. Most support 50–100+ languages; mixing languages on one page needs explicit config.
What You Can Do With an OCR'd PDF
Search
Ctrl+F / Cmd+F works. "Find all mentions of 'liability'" across 500 scanned contracts. This alone justifies OCR for archives.
Copy
Select text → copy → paste into Word, Excel, email. No retyping.
Edit
This is the tricky one. OCR creates a text layer behind the image. In Acrobat Pro, you can "Edit PDF" and modify that text layer — but the original image stays. You're editing a ghost. The visual page doesn't update unless you also edit the image (which Acrobat doesn't do well). For true visual editing, you need the source document or a full recreation.
Accessibility
Screen readers need text. OCR + tagged PDF = accessible scanned docs. Required for Section 508 / WCAG compliance.
Free OCR Options
Google Drive OCR (Free, Server-Side)
Upload PDF/image to Google Drive → Right-click → Open with → Google Docs. Docs creates a new document with the extracted text + the original images inline. Free, decent accuracy, supports 200+ languages.
Catch: Your document uploads to Google. Not for confidential data. Also: the output is a Google Doc, not a searchable PDF. You'd need to download as PDF to get a searchable PDF (which Docs does, but layout is reconstructed).
Online OCR Tools (Free Tiers, Server-Side)
Sites like onlineocr.net, ocr2edit.com, pdf2go.com — upload, get searchable PDF or text. Free tiers: few pages/day, file size limits.
Catch: Upload required. Privacy policies vary. Output quality varies.
Tesseract (Free, Open Source, Local)
The classic open-source OCR engine. Command-line, or via GUI frontends (gImageReader, OCRmyPDF). Runs 100% locally. No upload. Supports 100+ languages. Can produce searchable PDF (OCRmyPDF).
Catch: Technical setup. CLI/terminal required. No GUI by default. Accuracy good but behind commercial engines on difficult layouts.
Our Roadmap Note — Client-Side OCR
We're often asked: "Why no OCR in your toolkit?" Honest answer: OCR is computationally heavy. Neural network inference in WebAssembly is possible (Tesseract.js exists) but slow — 10–30 seconds per page on desktop, minutes on mobile. Battery drain. Memory pressure. We're watching Wasm SIMD and WebGPU — when client-side OCR becomes practical (sub-5-second/page on phone), we'll add it. For now: use the options above.
How to Get Better OCR Results
Scan at 300 DPI
This is the single biggest factor. Phone scanner apps (Adobe Scan, Microsoft Lens, iOS Notes, Google Drive) default to ~200–300 DPI. Check settings. Desktop scanners: set 300 DPI explicitly. Not 600 (diminishing returns, 4× file size). Not 150 (accuracy cliff).
Deskew
Crooked pages confuse layout analysis. Most scanner apps auto-deskew. Desktop scanners: enable "deskew" in driver. If you have a skewed PDF, our Rotate PDF tool can fix 90° rotations; for slight skews, you need an image editor or OCRmyPDF's --deskew flag.
Clean backgrounds
Remove coffee stains, hole-punch shadows, binder clips, sticky notes before scanning. If the scan exists: image editor → levels/curves → crush background to pure white, text to pure black. OCRmyPDF has --clean flag for this.
Black & white vs color
For text documents: scan B&W (1-bit) or grayscale. Color adds file size and can confuse OCR on colored text/backgrounds. Exception: you need the color images preserved (e.g., signed contracts with blue ink). Then scan color, run OCR, keep both layers.
Limitations
Handwriting
Standard OCR = printed text. Handwriting recognition (HTR) is a different, harder problem. Google Cloud Vision and Amazon Textract have HTR models — but they're paid, server-side, and accuracy on cursive is still 60–80%. For handwritten forms: transcribe manually.
Low-quality scans
Fax thermal paper, carbon copies, water damage, highlighter bleed, wrinkled pages — OCR degrades fast. No tool fixes "information not in the pixels."
Complex layouts
Newspaper columns, magazine layouts, forms with checkboxes, tables spanning pages — layout analysis (detecting reading order, columns, tables) is separate from character recognition. Commercial engines do it better. Tesseract needs config. Google Drive does okay.
Not "editable" in the visual sense
OCR gives you a text layer. You can search, copy, and in Acrobat Pro, edit that text layer. But the visual page (the scan image) doesn't change. You're editing a transparent overlay. For true visual editing, you need the source file or a redesign.
FAQ
How do I make a scanned PDF editable?
Run OCR. Free options: Google Drive (upload → Open with Docs), Tesseract/OCRmyPDF (local, CLI), online OCR sites (upload). Output: searchable PDF or text file. For visual editing, you still need the source or a desktop editor.
Is OCR free?
Yes — Google Drive, Tesseract, OCRmyPDF, and several online tools have free tiers. Commercial engines (Adobe, ABBYY, Google Cloud Vision API) are paid.
Can Google Docs OCR a PDF?
Yes. Drive → Upload PDF → Right-click → Open with → Google Docs. Creates a Doc with extracted text. Uploads to Google.
How accurate is OCR?
300 DPI, clean scan, standard font: 99%+. 150 DPI: ~80%. Handwriting: 60–80% (with HTR). Garbage in, garbage out.
How do I extract text from a scanned document?
OCR → copy text. Or OCRmyPDF with --output-type txt. Our PDF to JPG can convert pages to images first if you need to feed them to another OCR tool.
Can I OCR a PDF without uploading?
Yes — Tesseract / OCRmyPDF run locally. No upload. Our toolkit doesn't have OCR yet (client-side too heavy), but we're watching WebAssembly advances.
What's the best free OCR software?
For local/private: OCRmyPDF (wraps Tesseract, produces searchable PDF). For ease: Google Drive. For batch/automation: Tesseract CLI.
Working with scanned PDFs? Our toolkit handles the rest: JPG to PDF (rescan workflow), Compress PDF (OCR'd files get heavy), Watermark PDF (batch labeling), Merge PDF (combine OCR'd chapters). Free. Private. No upload.