Kraken

Historical documents and non-Latin scripts are the focus of Kraken, a turn key OCR system developed at the Ecole Pratique des Hautes Etudes in Paris. It provides fully trainable layout analysis, reading order detection, and character recognition, with support for right-to-left, bidirectional, and top-to-bottom writing such as Arabic and Hebrew. Recognition output comes in ALTO, PageXML, abbyyXML, and hOCR formats with word bounding boxes and character cuts. Pretrained models are shared through a public repository on Zenodo, and users can train their own segmentation and recognition models on custom material. Kraken installs via pip on Linux and macOS, including ARM machines, and pairs closely with eScriptorium, a web interface for annotating data, training models, and running transcription. Digital humanities researchers, libraries, and archives digitizing manuscripts and early printed books are its main users. Released under the Apache 2.0 license.