Research On Artistic Font Generation Based On Disentangled Glyph And Text Effect

With the advancement of modern technology,artistic fonts have emerged and followed the trend of the digital era,entering people’s vision.Artistic fonts have been widely used in the digital age and have become an indispensable design language in modern graphic design.The application of artistic fonts is very extensive and can be used in various graphic design works such as posters,advertisements,business cards,packaging,and brochures.Constructing artistic fonts is a unique creative technique for designers,who can adjust the line,shape,size,and color of fonts to make ordinary text more artistic and personalized.Carefully designed artistic fonts can endow text with great visual appeal,and these stunning effects often stem from the glyph and text effect style of the artistic font,which together reflect the designer’s ideas and emotions.Among them,the glyph is based on geometric structural features and exists in multiple ways,such as length and shortening deformation,straight and curved deformation,hook deformation,and directional deformation.Text effects are special effects based on character content and glyph,such as flame,metal,stripes,color gradients,etc.However,creating artistic fonts requires experienced designers who need to use professional editing software to complete the artistic font library.Therefore,the production of artistic fonts is both a creative and labor-intensive task.With the continuous development of deep learning and AIGC technology,automatic generation of artistic fonts has become a research field that receives considerable attention.Currently,some deep learning-based artistic font generation techniques have achieved good results.These techniques can automatically generate artistic font images with artistic sense and personalization by learning from a large number of artistic font samples.However,existing artistic font generation work still faces three problems:complex glyphs and delicate text effects are difficult to synthesize,high-quality and diverse artistic font datasets are lacking,and existing algorithms lack exploration of style disentanglement and recombination for artistic fonts.To address these problems,this paper mainly carries out the following three works:1)A baseline model DSE-Net for disentangle style encoding of the glyph style and text effect style of the artistic font is proposed to obtain the fine-grained features of the glyph and text effect,and improve the quality of generated artistic font;2)Develop a large-scale artistic font image dataset SSAF that supports style disentanglement research,with a total of 921,600 artistic font images.It contains standard,creative,handwritten,calligraphy Chinese artistic fonts,as well as 52 English alphabet artistic fonts;3)A compositional zero-shot artistic font synthesis model CAFS-GAN is proposed.This model constructs the style attribute space of the glyph and text effect through the style disentangled,and uses the composability of the glyph and text effect attribute to synthesize artistic font with a new integrated style.This paper proves the superiority of our proposed DSE-Net and CAFS-GAN through a large number of experiments,and the effectiveness of our proposed dataset in verifying the synthesis of glyphs and text effects.In addition,this paper has carried out engineering implementation of the above mentioned methods,and integrated the construction of artistic font image generation system--Art Palace.