Japanese historical character recognition using deep convolutional neural network (DCNN) with dropblock regularization

Abstract

The history of a country always helps in the development of making modern society. Japanese classical handwritten/literature dataset Kuzushiji-MNIST and Kuzushiji-49 recently introduced in public domain by Center of Open Data in the Humanities (CODH). The availability of the datasets invite the researchers in different domains. The hidden treasure of information in these datasets resolved and worked by many researchers. The research on these datasets can be helpful to explore and analyze the detail of Japanese era/literature, which ultimately helps in the development of modern society. In this study, Deep Convolutional Neural Network (DCNN) with DropBlock regularization is used to recognize the hiragana characters in Japanese historical script.