| 秦嘉霖,刘维尚.基于直观汉字构形原理的C3-GAN字体生成优化方法[J].包装工程,2023,44(10):193-201, 268. |
| 基于直观汉字构形原理的C3-GAN字体生成优化方法 |
| C3-GAN Fonts Generation Optimization Based on Intuitive Chinese Character Configuration |
| |
| DOI：10.19554/j.cnki.1001-3563.2023.10.019 |
| 中文关键词: 生成对抗网络 汉字构形 人工智能 深度学习 汉字字体 C3-GAN |
| 英文关键词: generative adversarial networks Chinese character configuration artificial intelligence deep learning Chinese character font C3-GAN |
| 基金项目:2023年河北省教育厅人文社会科学研究重大课题攻关项目(ZD202327)阶段性成果。 |
|
| 摘要点击次数: |
| 全文下载次数: |
| 中文摘要: |
| 目的 为了提升生成对抗网络汉字风格迁移的图像生成质量，实现汉字智能生成在字库产业中的实际应用，提出了一种基于直观汉字构形学的条件生成对抗网络字体生成优化方法(Optimization of Conditional Fonts Generation with Chinese Character Configuration GANs，C3-GAN)。方法 建构了直观汉字构形模组(C3 Module)，该模组包含了利于条件生成对抗网络进行汉字构形语义特征学习的全特征汉字字符集。C3-GAN在条件生成对抗网络模型下进行字体生成训练，降低了必要训练样本数量，实现对字体生成效果的优化。结果 使用C3-GAN生成汉字图像的清晰度更高、字形更准确。在图像相似性定量评估中，使用C3-GAN的实验组相比于其他模型，获得了更高的相似值和更小的误差值。结论 使用C3-GAN可以降低必要训练样本数量、提升汉字图像质量。在实际项目中具有一定的应用性和可操作性。 |
| 英文摘要: |
| The work aims to propose a method for Optimization of Conditional Fonts Generation with Chinese Character Configuration GANs (C3-GAN) of the intuitive Chinese character configuration to improve the image generation quality of Chinese character style transferring with generative adversarial networks, and achieve the practical application of Chinese character intelligent generation in the font industry. An intuitive Chinese character configuration module (C3 Module) was constructed, which contained Chinese character sets with all features. It was beneficial to generating an adversarial network for the learning process of semantic features of Chinese character configuration. Performing font generation training with C3-GAN under the model of the conditional generative adversarial network reduced the number of compulsory training samples, and optimized the font generation effect. C3-GAN could generate Chinese characters with higher images definition and more accurate glyphs. In the quantitative evaluation of image similarity, the experimental group using C3-GAN obtained higher similarity values and smaller error values than other models. C3-GAN can reduce the number of compulsory samples, and improve the image quality of Chinese characters. It has certain applicability and operability in practical projects. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
