Image to Text (OCR): Extracting Text from Photos and Scans
Optical Character Recognition (OCR) reads text from images β receipts, screenshots, scanned documents, signs in photos. Modern OCR is accurate enough for most everyday use, though handwriting and unusual fonts still trip it up.
OCR works by training a model to recognize letter shapes regardless of size, font, color, or background. Modern open-source engines (like Tesseract) reach 95%+ accuracy on clean printed text in well-lit scans. Handwriting, low-contrast captures, or non-Latin scripts are harder.
EllyTools' OCR runs entirely in your browser using Tesseract.js β no upload required.
What OCR is good at
- β’Printed pages with consistent fonts
- β’Screenshots of websites, error messages, code
- β’Receipts and business cards
- β’Books and PDFs that aren't already digitized
What it struggles with
- β’Cursive or sloppy handwriting
- β’Multi-column layouts where columns merge
- β’Heavy backgrounds bleeding into text
- β’Languages outside the trained set (some Asian scripts especially)
Extended FAQ
Does OCR work on Korean / Japanese / Chinese?
Tesseract supports CJK languages but accuracy is lower than for Latin scripts. Better results come from Korean-trained models specifically.
Can OCR read PDFs directly?
If the PDF was scanned (image-based), yes β extract images first, then OCR. If the PDF has digital text, just copy it directly.
Are my images uploaded?
No. The recognition engine runs in your browser, so the image never leaves the device. The internet is only needed the first time, to download the engine and the language data.