Attention U-Net Network for Chinese Calligraphy Style Generation

Chinese characters, owing to their structural traits, have highly variable writing styles. In order to achieve the digital inheritance and innovation of calligraphic Chinese characters and styles, this paper presents an Attention U-Net model integrating attention mechanisms into the U-Net for digital generation and transfer of Chinese calligraphy styles. The model extracts stylistic and structural features via an encoder, achieves the decoupled control of style and structure through dynamic fusion in multi-level feature spaces and learnable attention weights. Experiments validate its excellence: it can not only capture stroke rhythms and structural differences across calligraphic styles, generate high-quality characters with natural brushwork and stylistic consistency, but also improve the accuracy and robustness of style transfer while preserving traditional charm.