Getting Accurate Text from a Scan (OCR), Including Urdu and Arabic

What OCR does

Optical character recognition looks at the shapes in a picture and guesses which letters they are. It works best on printed text that is sharp, straight and dark on a light background. It struggles with handwriting, decorative fonts, text over patterns or photos, and very small print.

Urdu written in Nastaliq style is harder for OCR than Arabic Naskh or Latin text, because letters join and stack diagonally. Printed Naskh-style Urdu and Arabic are recognised much more reliably.

Prepare the page

1. Scan the page flat and straight first (Scan to PDF does this), so lines of text are horizontal.
2. Use B&W or Lighten so the background is white and the letters are dark.
3. Crop away anything that is not text you need, such as photos, logos and table borders, when possible.
4. Choose the right language before reading. Mixing up languages is the most common cause of nonsense output.

Check before you rely on it

Always read the recognised text against the original for names, numbers and dates. OCR confuses similar shapes: 0 and O, 1 and l, 5 and S, and in Arabic script letters that differ only by dots. For anything legal or financial, the original document is what counts.