Optical Character Recognition

Optical character recognition (OCR) is the technology that converts text appearing in images or scanned documents into machine-readable characters. A traditional pipeline detects text regions, segments them into lines and characters, classifies each shape, and then runs a language pass that corrects unlikely letter sequences. Deep learning replaced most hand-built stages and pushed accuracy far higher on messy inputs such as photographs, curved surfaces, handwriting, and mixed scripts. The output makes previously inert documents searchable, indexable, and usable by downstream automation. Tesseract remains the widely used open source engine, while Google Cloud Vision and AWS Textract offer hosted services, and Textract in particular extracts table and form structure rather than only raw characters. Multimodal models such as GPT and Claude now read text in images directly and can answer questions about a document in one step, blurring the line between OCR and document understanding. Typical uses include processing expenses from receipts, entering invoice data, and digitizing archives. Accuracy still depends heavily on input quality, since low resolution, skew, glare, and poor contrast produce errors that flow silently into downstream systems, so pipelines feeding accounting or legal records need confidence thresholds and human review.