You've received a scanned contract. You need to find a specific clause, copy an account number, or search for a date. You open the PDF and try to select text - and nothing happens.
That's because scanned PDFs aren't text documents. They're photographs of text, stored as images inside a PDF container. To a computer, the "words" on the page are just pixels arranged in shapes.
OCR - Optical Character Recognition - is the technology that bridges that gap.
How OCR Works
At a high level, OCR does three things:
-
Image preprocessing: The raw scanned image is cleaned up - deskewed (rotation corrected), despeckled (noise removed), contrast-enhanced. A 1% improvement in preprocessing quality translates to significantly better recognition accuracy.
-
Character recognition: Machine learning models analyze the shapes of individual characters and map them to text. Modern OCR engines (like Tesseract, Google Cloud Vision, and AWS Textract) use neural networks trained on millions of document images.
-
Layout reconstruction: The engine attempts to reconstruct the logical structure - paragraphs, columns, tables, headers - from the spatial arrangement of recognized text on the page.
The output is a text layer added to the PDF. The image stays visible; the text layer is invisible but selectable, searchable, and copyable.
When You Need OCR
Not every PDF needs OCR. A PDF created digitally - exported from Word, generated by a browser print function, or created by design software - already contains real text. You can select and search it immediately.
You need OCR when:
- The PDF was created by scanning a paper document. Medical records, legal filings, historical archives, signed contracts - anything that started on paper.
- The PDF is an "image PDF." Some tools (certain printers, fax-to-PDF services) create PDFs where even digitally-created content is converted to images.
- You receive documents from older systems that exported image-only PDFs.
An easy test: try selecting text in your PDF viewer. If you can't select any characters, it's image-only and needs OCR.
Accuracy Factors
OCR accuracy varies widely. Here's what affects it:
Scan quality
This is the single biggest factor. A sharp, well-lit, high-contrast scan (300 DPI or higher) will yield near-perfect results. A dark, skewed, low-resolution scan will produce garbled output.
If you're in control of the scanning process:
- Use 300 DPI minimum (600 DPI for fine print)
- Ensure even lighting without shadows across the page
- Keep the scanner glass clean
- Use black-and-white mode for text-only documents (smaller file, often better contrast)
Font and typeface
Clear, common serif and sans-serif fonts recognize extremely well. Decorative fonts, handwriting, very small print, and condensed typefaces are harder.
Language
Most OCR engines default to English. If your document is in another language, select the correct language model - accuracy can drop significantly without it.
Document condition
Yellowed paper, faded ink, handwritten annotations, and physical damage all reduce accuracy. There's a floor on how well software can recover from a genuinely poor source document.
What to Do With Low-Confidence Results
Even good OCR occasionally makes mistakes. For professional use, you should spot-check the output:
- Search for key terms you expect to find
- Check numbers and dates specifically - these are commonly misread
- Look at areas with unusual formatting, headers, and footers
For critical documents (contracts, financial records), a human review pass is worth the time.
OCR for Search vs OCR for Editing
There are two levels of what you can do with OCR output:
Searchable PDF (most common): The text layer is added to the existing scanned image. The document looks identical, but you can now search, select, and copy text. This is what SupaPDF produces by default.
Fully converted text document: The OCR output is used to reconstruct a Word or plain text document. This is useful if you need to edit the content - but formatting reconstruction is imperfect, and complex layouts (tables, columns, mixed content) often need manual cleanup.
Running OCR with SupaPDF
- Go to OCR PDF
- Upload your scanned PDF
- Select the primary language of the document
- Process - SupaPDF adds a searchable text layer while keeping the original appearance intact
- Download the OCR'd PDF
For documents with multiple languages on the same page, select the dominant language for best results. Multi-language support within a single document pass is available on Premium.
After OCR: What's Possible
Once a scanned PDF has a proper text layer, it becomes a full participant in your document workflow:
- Search across the document and within document management systems
- Copy text for use in other applications
- Convert to Word or Excel for editing (use PDF to Word)
- Redact sensitive content properly (image-based PDFs can't be truly redacted without OCR first)
- Extract data programmatically via the SupaPDF API
OCR is often the first step in a larger workflow. A batch of scanned invoices becomes searchable, then extractable, then importable into your accounting system - all because the text layer was added.
The technology isn't perfect, but for well-scanned documents in standard languages, modern OCR gets you 99%+ accuracy. That's good enough to be genuinely useful.