Abstract:Synthesizing Chinese characters with consistent style using few stylized examples is challenging. Existing models struggle to generate arbitrary style characters with limited examples. In this paper, we propose the Generalized W-Net, a novel class of W-shaped architectures that addresses this. By incorporating Adaptive Instance Normalization and introducing multi-content, our approach can synthesize Chinese characters in any desired style, even with limited examples. It handles seen and unseen styles during training and can generate new character contents. Experimental results demonstrate the effectiveness of our approach.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **Generate Chinese characters with arbitrary styles based on a few - shot (few - sample examples), while maintaining the consistency of character content**. Specifically, existing models have difficulty in generating diverse stylized characters when using limited examples, especially when dealing with complex Chinese characters. The Generalized W - Net proposed in this paper aims to overcome these problems by introducing new architectures and improved techniques. ### Problem Background 1. **Complex Character Structure**: Chinese characters are composed of multiple strokes and radicals, with complex and diverse structures. 2. **Difficult Style Transfer**: Existing deep - generation models have poor performance when dealing with handwritten characters, especially when there are only a few examples. 3. **Separation of Style and Content**: A method is required to separate the content and style of characters and be able to flexibly switch between different styles. ### Solutions To solve the above problems, the paper proposes the following improvements: 1. **Extension of the W - Net Architecture**: - **Multiple Content Prototypes**: Introduce multiple content prototypes so that the model can combine different style features. - **Adaptive Instance Normalization (AdaIN)**: Used to enhance the effect of style transfer. - **Residual Blocks and Dense Blocks**: Improve the performance of the model. 2. **Feature Mixer**: - Use the AdaIN technique to normalize the extracted content features, making them better adapt to different style features. - Introduce multiple normalization methods in the feature mixer, such as Batch Normalization (BN), Instance Normalization (IN), and Layer Normalization (LN), to improve the robustness and generalization ability of the model. 3. **Adversarial Training**: - Train based on the Wasserstein GAN with Gradient Penalty (W - GAN - GP) framework to ensure that the generated characters are visually close to real characters. - Introduce multiple loss functions, including adversarial loss, pixel reconstruction loss, and perceptual loss, to optimize the generation effect. ### Experimental Results The experimental results show that the Generalized W - Net can not only generate high - quality stylized characters with a few examples, but also handle unseen styles and generate new character content. In addition, the model also demonstrates cross - language generalization ability and can generate characters of other East Asian languages (such as Korean and Japanese). ### Summary By proposing the Generalized W - Net, this paper solves the problem of existing models in generating diverse stylized Chinese characters in the few - sample case, providing new ideas and technical means for future research.

Generalized W-Net: Arbitrary-style Chinese Character Synthesization

W-Net: One-Shot Arbitrary-Style Chinese Character Generation with Deep Neural Networks

Handwritten Chinese Font Generation with Collaborative Stroke Refinement.

FontGAN: A Unified Generative Framework for Chinese Character Stylization and De-stylization

Breaking the Representation Bottleneck of Chinese Characters: Neural Machine Translation with Stroke Sequence Modeling

VCWE: Visual Character-Enhanced Word Embeddings

An end-to-end chinese font generation network with stroke semantics and deformable attention skip-connection

GAN-Based Unpaired Chinese Character Image Translation Via Skeleton Transformation and Stroke Rendering.

Deep stacking network-based structure information guided Chinese character library generation method

Visual Attention Adversarial Networks for Chinese Font Translation

Separating Style and Content for Generalized Style Transfer

ZiGAN: Fine-grained Chinese Calligraphy Font Generation via a Few-shot Style Transfer Approach

CalliGAN: Style and Structure-aware Chinese Calligraphy Character Generator

A Novel Three-Staged Generative Model for Skeletonizing Chinese Characters with Versatile Styles

Style learning-based Chinese character synthesis method

SCFont: Structure-Guided Chinese Font Generation Via Deep Stacked Networks

Artistic Glyph Image Synthesis via One-Stage Few-Shot Learning

Learning to Write Stylized Chinese Characters by Reading a Handful of Examples

Auto-Encoder Guided GAN for Chinese Calligraphy Synthesis

Decoupled Representation Learning for Character Glyph Synthesis

Chinese character style transfer based on multi-scale GAN