Identification of Scribes from Historical Manuscripts

Abstract

Computerized analysis of historical documents has remained an interesting research area for the pattern classification community for many decades. From the perspective of computerized analysis, key challenges in the historical manuscripts include automatic transcription, dating, retrieval, classification of writing styles and identification of scribes etc. The focus of our current study lies on identification of writers from the digitized historical manuscripts. The documents are first pre-processed to segment handwriting from the background. For feature extraction and subsequent classification, we extract small patches of handwriting. These patches are extracted in two different ways, by a dense sampling of handwriting using small windows as well as by finding the key points in handwriting and using these key points as centers of small windows to extract writing fragments. Features are extracted from writing windows using a two-step fine-tuning of convolutional neural networks. First, the ConvNets are trained on contemporary handwriting samples and then fine-tuned to the limited set of historical manuscripts (Papyrus). Decisions on patches are combined using a majority vote to decide the authorship of a query document. Preliminary experiments on a set of challenging and degraded manuscripts report promising performance