Tesseract

Optical character recognition here comes as a C and C++ library, libtesseract, plus a tesseract command-line program that turns scanned pages and photos into machine-readable text. Since version 4 the default recognizer is an LSTM network that reads whole text lines, while the older pattern-matching engine from the 3.x era remains selectable. Languages ship as separate traineddata sets: fast integer models, slower float models that also serve as the base for fine-tuning, and a combined set that keeps the legacy models, covering well over 100 languages and scripts. It reads PNG, JPEG, TIFF and other images through Leptonica and writes plain text, hOCR, searchable PDF, TSV, ALTO or PAGE XML. It runs on an ordinary CPU, installs from apt, Homebrew, MacPorts or the UB Mannheim Windows installers, and reaches Python and other languages through third-party wrappers; there is no GUI. Begun at HP and developed by Google until 2017, it is now community maintained under Apache 2.0, which permits commercial use. A sensible default for developers who need offline OCR without per-page fees.