Artistic glyph image synthesis via one-stage few-shot learning

Abstract

Fig. 1. (a) An overview of our method, given a few reference samples (5 for English or 30 for Chinese), glyph images of all other characters in the same style can be synthesized. (b) Examples of synthesized English/Chinese glyph images obtained by our proposed AGIS-Net, MC-GAN [Azadi et al. 2018] and TET-GAN [Yang et al. 2019], respectively, please zoom in for better inspection.Automatic generation of artistic glyph images is a challenging task that attracts many research interests. Previous methods either are specifically designed for shape synthesis or focus on texture transfer. In this paper, we propose a novel model, AGIS-Net, to transfer both shape and texture styles in one-stage with only a few stylized samples. To achieve this goal, we first disentangle the representations for content and style by using two encoders, ensuring the multi-content and multi-style generation. Then we utilize two collaboratively working decoders to generate the glyph shape image and its texture image simultaneously. In addition, we introduce a local texture refinement loss to further improve the quality of the synthesized textures. In this manner, our one-stage model is much more efficient and effective than other multi-stage stacked methods. We also propose a large-scale dataset with Chinese glyph images in various shape and texture styles, rendered from 35 professional-designed artistic fonts with 7,326 characters and 2,460 synthetic artistic fonts with 639 characters, to validate the effectiveness and extendability of our method. Extensive experiments on both English and Chinese artistic glyph image datasets demonstrate the superiority of our model in generating high-quality stylized glyph images against other state-of-the-art methods.

Write Like You: Synthesizing Your Cursive Online Chinese Handwriting via Metric‐based Meta Learning

“…Recently, lots of CNN‐based models for offline Chinese glyph synthesis have emerged [RMC15, SRL*17, ZPIE17, Tia17, ZZC18, GGL*19, WGL20]. However, these methods fail to reflect the process of human writing, and cannot handle scribbled handwritings (see Figure 1).…”

MLFont: Few-Shot Chinese Font Generation via Deep Meta-Learning

“…Multi-ContentGAN [3] is the first method for the English artistic font generation, which cannot handle the Chinese characters. Both TET-GAN [35] and AGIS-Net [6] are proposed for the Chinese artistic font generation. Existing text effects transfer methods focus on the transfer of texture styles, and cannot cope with the task of font generation for geometric style transfer.…”

JointFontGAN

“…MC-GAN [1] presents a stacked conditional GAN (cGAN) architecture to predict the coarse glyph shapes, and an ornamentation network to predict color and texture of the final glyphs. Unlike two-stage MC-GAN framework, AGIS-Net [8] transfers both shape and texture styles in one-stage with only a few stylized samples in order to improve the computational efficiency. FontRNN [24] treats Chinese characters as sequences of points (writing trajectories) and proposes to handle the font generation task via a Recurrent Neural Network (RNN) model.…”

“…Recently, data-driven approaches have been proposed to robustly investigate the correlations between different objects / instances without relying on hard-coded metrics. Along this direction, there is an emerging trend to automatically generate different styles of font / glyph by deep neural network (DNN), such as DNN-based method [3], modified variational autoencoder (VAE) method [26], image-to-image translation model [18], zi2zi [25], DCFont [12], Hierarchical Adversarial Network (HAN) [7], CycleGAN-based model [6], SCFont [13], PEGAN [22], MC-GAN [1], AGIS-Net [8], FontRNN [24], convolutional recurrent generative model [4], Glyph-GAN [10], FontGAN [16], etc. However, most of the above methods are not able to automatically generate large-scale fonts with highquality and high-consistency in geometry and content / style for characters and words / paragraphs from a few given samples.…”