Macarius: A HTR Model for Romanian Slavonic Early Printed Books

The paper describes the process of creating and evaluating the HTR (Handwritten Text Recognition) model for Romanian Slavonic early printed books (first half of the 16th century, Middle Bulgarian Church Slavonic, Cyrillic Script) using the Transkribus software platform, based on the principles of artificial intelligence, machine learning and advanced neural networks. The HTR model was created on the material of Romanian Slavonic early printed books from Târgovişte printing house: the Liturgikon from 1508 and the Teatraevangelion from 1512 from the oldest printing house managed by hieromonk Macarius, as well as the Apostle from 1547 from the printing house managed by Dimitrije Ljubavić. The most important result of the paper is the creation of the first version of the generic HTR model Macarius (named in honour of hieromonk Makarije, the first South Slavonic and Romanian printer) with exceptional performance – the percentage of incorrectly recognized characters (including accent marks) is only 2.7%. Research has shown that this HTR model can also be used for the automatic recognition of Romanian Slavonic early printed books published in the second half of the 16th century. HTR model Macarius together with Ground Truth data is available to all users of the Transkribus platform, which ensures its wider use, as well as the possibility for further improvement of its performance.