A Hybrid Vision Transformer and Convolutional Neural Network for Chinese Calligraphy Style Classification

Chinese calligraphy, a highly regarded art form with a history spanning thousands of years, is considered a valuable cultural heritage. Accurately identifying unique calligraphy styles is essential for art collection, authentication, and academic research. In this paper, we propose a novel hybrid architecture combining convolutional neural networks (CNNs) and Vision Transformer (ViT) for the classification of calligraphy styles. Our model incorporates a Calligraphic Feature Attention (CFA) module to enhance sensitivity to distinct stroke features, such as direction, thickness, and spacing, which are crucial for distinguishing between different calligraphers. We constructed a new dataset containing over 6700 samples from five renowned historical Chinese calligraphers: Huang Tingjian, Liu Gongquan, Wang Xizhi, Song Huizong, and Yan Zhenqing. This dataset was used to train and evaluate our proposed method. Comparative experiments with classic models like ShuffleNet, DenseNet, EfficientNet, and MobileViT show that our approach achieves superior performance, with an accuracy of 97.4%, outperforming all other methods. Our results demonstrate the potential of combining CNN's local feature extraction with Transformer's global context understanding to effectively classify intricate Chinese calligraphy styles.