A Chinese Font Generation Method based on Dynamic Convolution Improved Generative Adversarial Network

Font generation is a challenging research topic in the field of computer vision, aiming to apply the style from a reference image to a source character image. The core task of font generation is to generate images, especially complex character images, that closely resemble the style of the reference font while maintaining the original glyph structure. This paper proposes a novel font generation method based on the generation adversarial network (GAN). Firstly, a new unsupervised end-to-end Chinese font generation network is designed. Then, a Convolution-Transformer (CNNTrans) module is devised to extract style features from global and local with attention constraint. Finally, a dynamic convolutional skip connection module (DCSC) is proposed to dynamically fuse font content and style according the importance of content and style features in different positions. Experimental results demonstrate that the proposed font generation method can generate higher quality styled fonts compared to other methods.