Location of Simple Graphemes in Mediaeval Manuscripts based on Mask R-CNN

In this paper we describe a system for the location of simple graphemes in mediaeval manuscripts based on the Mask R-CNN convolutional neural network. This is the first step towards the ambitious goal of providing palaeographers with a powerful tool with which to speed up and refine the delicate process of dating and determining the origin of manuscripts. In order to train the network, a new dataset composed of49 pages of Latin Middle Ages manuscripts has been built. Experimental results demonstrate that using the Mask R-CNN network, along with a proper configuration of parameters, leads to good overall outcomes of classification.