What Is OCR? Optical Character Recognition Explained
OCR converts scanned documents and images into searchable, editable text. Here's how it works, how accurate it is, and when to use it.
Advertisement
OCR — Optical Character Recognition — is the technology that converts images of text (scanned documents, photos of signs, screenshots) into machine-readable text that you can search, copy, edit, and process. Without OCR, a scanned PDF is just a photograph — Google can't index it, you can't Ctrl+F through it, and you can't copy text from it.
What Is OCR?
OCR was first developed in the 1970s and has improved dramatically with machine learning. Modern OCR engines — including open-source Tesseract (developed by HP, now maintained by Google) — can recognise text in 100+ languages, handle fonts, handle italics and bold, and even recognise text at various angles and with moderate distortion.
How OCR Works
- Pre-processing — deskewing, denoising, and binarising the image to increase contrast
- Segmentation — identifying page regions: columns, paragraphs, lines, words, characters
- Character recognition — matching each segment against a trained character model
- Post-processing — applying language model to correct likely recognition errors
- Output — structured text with optional position data (bounding boxes)
OCR Accuracy and Factors That Affect It
Clean, professionally-scanned documents in English typically achieve 97–99% character accuracy with modern engines. That means fewer than 3 errors per 100 characters — excellent for most purposes. Accuracy drops with: low scan resolution (under 150 DPI), skewed or tilted pages, handwriting, decorative fonts, faded ink, and complex multi-column layouts.
Scan at 300 DPI for best results
The single most impactful variable is scan resolution. Scanning at 300 DPI gives OCR engines enough pixel detail to distinguish similar characters (like l, 1, and I). Most scanners default to 200–300 DPI; smartphone scan apps often default to lower resolutions — check your settings.
Common OCR Use Cases
- Digitising paper records and archives into searchable databases
- Extracting data from invoices, receipts, and purchase orders for accounting
- Making scanned legal documents searchable and citable
- Converting physical books and research papers to digital text
- Accessibility — making scanned PDFs readable by screen readers
- Extracting text from screenshots for editing or translation
Extract text from your PDF
Our free OCR tool supports 100+ languages and processes everything in your browser.
OCR Limitations
- Handwriting — OCR is trained on printed text and performs poorly on handwriting
- Complex tables — multi-row merged cells often lose structure during OCR
- Mathematical equations — symbols are frequently misrecognised
- Non-Latin scripts — quality varies by language; Latin-script languages are best supported
- Very small text (under 8pt in the original) — insufficient pixel data for reliable recognition
Frequently Asked Questions
Can OCR read handwriting?
Standard OCR is designed for printed text. Handwriting OCR (ICR — Intelligent Character Recognition) is a separate, harder problem. Browser-based OCR tools generally do not support handwriting. Dedicated handwriting recognition is available in some premium tools and mobile apps.
Is OCR processing private?
When using our browser-based OCR PDF tool, all processing runs locally using WebAssembly — your document never leaves your device. This is important for sensitive documents like financial records or confidential contracts.
What resolution should I scan at for OCR?
300 DPI is the recommended minimum for good OCR accuracy. For documents with very small text (footnotes, captions, small-print legal text), scan at 400–600 DPI. Higher resolution increases file size but also OCR accuracy.