查看文章

Writer identification remains a very interesting challenge where many researchers have tried to find the various parameters which can help to find the right writer of a handwritten text. The introduction of deep learning has made it possible to achieve unprecedented records in the field. However, the question to ask, what size of patch to use to train a CNN model in order to have the best performance? In our paper, we try to find an answer to this question by investigating the results of the use of several patch sizes for a Resnet34 model and two languages Arabic and French from the LAMIS-MSHD dataset.