Automated writer identification for Syriac scribes

Several computer-based methods of writer-identi cation for the Syriac language text are presented, along with an analysis of their relative e ectiveness. Our aim is to implement various writer- identi cation methods in Syriac for scholars of the language and culture to use for academic pur- poses. Though these algorithms have been implemented for modern languages such as English, none have been used for Syriac. Possible applications of such a system include identifying a large num- ber of works potentially by a similar scribe, thus providing his possible oeuvre, and determining the author of a document given only a few pages. For in-document recognition, a combination of text-dependent and text-independent techniques was able to reach 100% accuracy with the data set. These results show that some of the methods implemented may be capable of providing reasonable accuracy in an actual scribe identi cation system. The contribution of this thesis is the initial stage in development of the writer identi cation system that will eventually be provided to the Syriac scholar community. This thesis also includes a brief analysis of the validity of di erent modern handwriting iden- ti cation methods for use with historical texts. Although the computational complexity is high, it was found that text-dependent methods based on the Congealing process outperformed simpler text-independent methods which perform well with modern handwriting