OCR (optical character recognition) is the process of converting an image of text — a scanned page, a photo of a sign — into actual machine-readable, selectable text characters.
From pixels to shapes
OCR software first identifies regions of an image that contain text, then breaks those regions down into individual character shapes, distinguishing letters and numbers from background noise, lines, or images.
From shapes to characters
Modern OCR uses machine learning models trained on enormous datasets of text in many fonts and handwriting styles to match each detected shape to the character it most likely represents, then assembles the results back into words and lines.
Why context matters
Good OCR systems also use language models to catch and correct likely errors — if a character is ambiguous between 'O' and '0', surrounding context (is it inside a word or a number sequence) helps resolve it correctly.
Does OCR accuracy vary by language?
Yes — OCR systems are typically trained per-language or per-script, so accuracy depends on how well-represented a given language and font style are in the training data; major languages in common fonts generally see the highest accuracy.