Ground Truth transcriptions for training OCR of historical Arabic handwritten texts
Add to collection
You do not have access to any existing collections. You may create a new collection.
This dataset comprises 120 digitised images (TIFF files) drawn from a selection of historical Arabic scientific manuscripts (10th-19th century) digitised through the British Library Qatar Foundation Partnership. Also contained are ground truth transcriptions (XML) for each page that can be used for training optical character recognition (OCR) or handwritten text recognition (HTR) software on historical Arabic handwritten texts. The folder contains the images and ground truth used for the RASM2018 and RASM2019 competitions, part of ICFHR 2018 and ICDAR 2019. The images are out of copyright and the transcriptions are also public domain.
The two .ZIP files are 3.73 and 4.96 GB, each containing 60 high resolution (400dpi) images in TIFF format accompanied by transcriptions in PAGE XML. Total download size for both parts (120 images and transcriptions) is 8.7 GB. Shelfmarks are used for all file names.