Text-Independent Writer Identification using Convolutional Neural Networks

摘要

This paper proposes an end-to-end deep-learning based method for text independent writer identification. In this method, convolutional neural networks (CNNs) are trained initially in order to extract features which represent characteristics of individual handwriting in the whole and in sub-regions of character images. During evaluating process, the extracted features from one or more characters are combined and used for identifying the writer. We apply a random sampling method to create a large number of training patterns required by CNNs during the training process. Experiments on the JEITA-HP database of Japanese handwritten character patterns show the effectiveness of this approach to overcome the difficulties of gathering handwritten character patterns of the same categories as the specimens of the writer.