All terms
Multimodal
Optical Character Recognition
Reading text out of images, such as scanned documents or photos.
Definition
Optical character recognition detects and pulls text out of images, scanned documents, or photographs. Early systems matched letters against fixed templates, while modern ones use neural networks—and increasingly vision-language models—to handle stylized, faint, or handwritten text and complex layouts. OCR is a key step in document processing, in RAG (fetching documents to answer questions) over PDFs, and in reading text inside multimodal inputs.