Image to Text (OCR)

OCR works best on sharp, evenly lit pictures where the text is reasonably large. These simple steps make the biggest difference.

Hindi and Marathi (both written in Devanagari), Bengali, Tamil, Telugu and Gujarati each have their own trained model, so choose the language of the text rather than just the script. Documents that mix English and an Indian language, such as forms, bills or certificates, read best with both selected, for example English + Hindi. Printed text works well; handwriting is generally not recognised reliably by Tesseract in any language.

Receipts are often printed in small, faint letters on curved paper: flatten the receipt, photograph it in good light and select only the lines you need. Screenshots are usually the easiest, because the letters are perfectly sharp; tiny ones are enlarged before reading. For multi-page scanned PDFs, use PDF OCR instead, which adds an invisible, searchable text layer to the PDF itself.

The first time you read text, your browser downloads the Tesseract engine (about 4 MB) and each chosen language (up to 3 MB) from this site and keeps them in its cache. Reading a page then takes a few seconds on a computer and a little longer on a phone. The confidence score is Tesseract's own estimate of how sure it is; below about 60% expect mistakes. Always check important numbers, such as amounts and IDs, against the picture.