Artistic font generation network combining font style and glyph structure discriminators

Artistic font plays an important practical value in advertising media and graphic design. The font is rendered to present a unique text effect, which is more ornamental and attractive. In order to explore more efficient font design method, this paper proposes artistic font generation network combining font style and glyph structure discriminators (ArtFontNet). We adopt the idea of generative adversarial network to obtain artistic fonts. The generator uses residual dense module to generate artistic fonts, and font style discriminator and glyph structure discriminator guide the generator together. The font style discriminator supervises the color and texture information of the entire font image. The glyph structure discriminator extracts the glyph structure and texture distribution of the generated image through the Canny edge detection operator. For the task of artistic font generation, our approach achieves significant performance compared to other existing methods. Compared with the font generation methods in the experiment, the PSNR value of the generated image in this paper is increased by 2.95 dB on average. The SSIM value is increased by 0.03 on average. The VIF value improved by 0.025 on average. The UV quantization results are maintained at 85%-90%. From both visual and objective evaluations, ArtFontNet enhances the detail fidelity and style accuracy of the generated artistic fonts.