Categorizing Chinese Glyphs Based on Liushu With Convolutional Neural Networks

As one of the most polupar languages in the world and the only live language of ideograph, Chinese can be learned and mastered fast and easily if the learners have the knowledge of Chinese glyphs evolution, categorization and creation detailed in Liushu — the earliest systematic theory on the structure of Chinese characters. However, there is no valid and accurate way from machine learning to classify Chinese glyphs based on Liushu. In this paper, we used Convolutional Neural Networks to recognize and classify images of Chinese glyphs based on Liushu and achieved 94% accuracy in the Duti/Heyi binary classification, which can be used for classifying unclassified Chinese glyphs and assisting Chinese language education.