Stroke-Based Few-Shot Chinese Character Style Transfer

Few-shot Chinese character style transfer aims to learn the style of Chinese characters with limited references and then generate a complete set of Chinese character fonts. This not only solves the repetitive and cumbersome manual effort in traditional Chinese character font design but also addresses the design challenges posed by the large number and complex structure of Chinese characters. Even though current methods have achieved some success with the support of large Chinese character font datasets, Few-shot Chinese character style transfer remains challenging due to the variations in the same stroke within the same font style. In this paper, we propose a novel method for Chinese character style transfer. Firstly, it reorganizes each Chinese character based on the 32 basic strokes of the reference font, capturing the local detailed features of the characters. Then, a cross-attention module is used to refine the stroke details of the character while learning the overall glyph structure of Chinese characters. Through this design, our model achieves good results on smaller datasets, demonstrating its superiority compared to other methods.