⌘K

Documents

OCR

OCR (Optical Character Recognition) converts text inside images and scanned documents into real, selectable and searchable characters.

Updated 12 Aug 2026

From pixels to characters

A scan looks like a document but contains no text — only a picture of one. That is why searching a scanned PDF returns nothing. OCR analyses the image, locates lines and glyphs, classifies each shape as a character, and outputs text that can be copied, searched and indexed.

  1. 1Pre-processing: deskew, straighten and increase contrast
  2. 2Binarisation: separate ink from background
  3. 3Layout analysis: find columns, paragraphs, lines and tables
  4. 4Recognition: classify glyphs, historically by template matching, now with neural networks
  5. 5Post-processing: apply a dictionary and language model to fix likely misreads

What drives accuracy

FactorEffect
Resolution300 DPI is the practical minimum for reliable results
Contrast and lightingShadows and grey backgrounds cause dropped characters
SkewEven a few degrees of rotation hurts line detection
TypefaceClean printed serif and sans text beats decorative or handwritten script
LanguageRecognition must be told the correct language and script

OCR output is never guaranteed correct. Digits and lookalike characters — 0/O, 1/l, 5/S — are the usual failures, so verify anything financial or legal before relying on it.

Compress PDF

Compress, merge and split scanned documents privately in the browser.

Try Compress PDF

Characteristics

  • Turns images of text into machine-readable characters
  • Modern engines use neural networks rather than template matching
  • Can produce a searchable PDF by layering invisible text over the scan
  • Accuracy is probabilistic and depends heavily on input quality

Common uses

  • Making scanned archives searchable
  • Extracting data from invoices, receipts and forms
  • Digitising books and historical records
  • Reading text from photographs for accessibility tools

Advantages

  • Unlocks decades of paper archives for search
  • Removes manual re-typing from data entry workflows
  • Improves accessibility for screen-reader users
  • Enables automated indexing and classification

Limitations

  • Accuracy collapses on low-resolution or skewed scans
  • Handwriting recognition remains unreliable
  • Complex tables and multi-column layouts are often mangled
  • Output needs human review for critical data

Examples

A low-quality scan producing classic confusions
Printed:    Invoice #10583 — Total £1,240.00
OCR output: lnvoice #IO583 — Total £1,24O.00
                ^ l for I     ^ O for 0

Frequently asked questions

What is OCR used for?

Turning scans and photographs of text into searchable, copyable characters for archives, invoices and forms.

Why can't I search my scanned PDF?

It contains images rather than text. Running OCR adds a text layer that search can index.

How accurate is OCR?

Well over 98% on clean 300 DPI printed text, and far lower on poor scans or handwriting.

Does OCR work on handwriting?

Only partially. Handwriting recognition is a harder, separate problem with much lower accuracy.

How can I improve OCR results?

Scan at 300 DPI or higher, straighten the page, raise contrast and select the correct language.

Related terms

All terms

Related tools

Related guides

All guides