書法是我國文字書寫的藝術表現，有豐富的形狀和變化。本論文嘗試使用生成對抗網路來模擬生成歷代名家的書法字體。我們以書法名家的書法字體影像為學習的訓練資料集。本研究中，我們以Zi2Zi的方法為基礎，實作書法字體風格轉換的生成對抗網路模型，並探討其他研究的一些方法對模型產生的影響，嘗試得出最佳化的深度學習模型。我們探討加入U-Net、類別內嵌(Category Embedding)、和HAN等方法，以及訓練資料集大小對模型造成的影響。在最後實驗中，我們比較了pix2pix模型、Zi2Zi模型、HAN模型、和HAN-Zi2Zi模型。經過實驗後發現，加入U-Net和Category Embedding都對模型的成果有所幫助，而使用越多字體進行訓練會有越好的成效。另外，HAN-Zi2Zi效果最好。

Calligraphy is the artistic expression of Chinese character writing, with rich shapes and variations. This research attempts to use the Generative Adversarial Network to generate the calligraphy characters of famous calligraphers. We use calligraphy font images of famous calligraphy masters as training materials for learning. In this research, based on the Zi2Zi method, we implemented a generational confrontation network model for calligraphic font style conversion, and explored the impact of some other research methods on the model, and tried to arrive at an optimized deep learning model. We discuss methods such as adding U-Net, Category Embedding, and HAN, as well as the impact of the size of the training data set on the model. In the final experiment, we compared the pix2pix model, Zi2Zi model, HAN model, and HAN-Zi2Zi model. After experimentation, it is found that adding U-Net and Category Embedding are helpful to the results of the model, and the more fonts are used for training, the better the results will be. In addition, HAN-Zi2Zi works best.

目錄
中文摘要	i
英文摘要	ii
目錄	iii
圖目錄	vi
表目錄	viii
第一章 緒論	1
1.1前言	1
1.2研究動機與目的	1
1.3論文架構	2
第二章 文獻探討	3
2.1圖像轉換	3
2.1.1生成對抗網路	3
2.1.2 Pix2Pix	5
2.1.3 CycleGAN	8
2.2 Zi2Zi漢字生成法	10
2.3 AEGG漢字生成法	12
2.4 HAN漢字生成法	12
2.5 CalliGAN漢字生成法	14
第三章 研究方法	16
3.1 資料前處理	18
3.2 生成網路	20
3.3 鑑別網路	24
3.4 損失函數	26
3.5 參數設定	27
第四章 實驗結果與分析	29
4.1 實驗環境	30
4.2 訓練資料大小比較	31
4.3 實驗評分指標	32
4.3.1 結構相似性指數評分	32
4.3.2 主觀評分	35
4.4 實驗結果與CalliGAN比較	36
第五章 結論與未來展望	37
參考文獻	38
附錄一 英文論文	41

 
圖目錄
圖 1  GAN架構圖	4
圖 2  U-Net架構圖	6
圖 3  Pix2pix架構圖	6
圖 4  PatchGAN圖形表示	7
圖 5  CycleGAN概念圖	8
圖 6  CycleGAN架構圖	9
圖 7  Zi2Zi架構圖	11
圖 8  AEGG架構圖	12
圖 9  部首比較	13
圖 10  HAN架構圖	13
圖 11  CalliGAN架構圖	15
圖 12  實驗介紹章節	16
圖 13  源字體（仿宋）以及7種字體	17
圖 14  實驗流程圖	18
圖 15  圖像前處理	19
圖 16  訓練資料合併	20
圖 17  生成網路流程	21
圖 18  鑑別器流程	24
圖 19  PatchGAN架構圖	25
圖 20  陳忠建書寫的7種字體	29
圖 21  各實驗結果比較	34
圖 22  與AEGG、CalliGAN比較結果圖	36


表目錄
表 1  訓練資料個數	19
表 2  Encoder詳細架構	22
表 3  Decoder詳細架構	23
表 4  Discriminator架構	25
表 5  參數設定	28
表 6  訓練資料數	30
表 7  實驗環境表	30
表 8  訓練資料大小比較	31
表 9  SSIM比較結果數據	34
表 10 主觀比較結果數據	35

參考文獻
[1]	Augustus Odena, Christopher Olah, and Jonathon Shlens. “Conditional image synthesis with auxiliary classifier GANs.” In ICML, 2017.

[2]	A. Radford, L. Metz, and S. Chintala. “Unsupervised representation learning with deep convolutional generative adversarial networks.” Computer Science, 2015.

[3]	Bart van Merrienboer Caglar Gulcehre, Dzmitry Bahdanau, Fethi Bougares Holger Schwenk and Yoshua Bengio. “Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation.” Computation and Language, 2014.

[4]	C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V. Vanhoucke, and A. Rabinovich. “Going deeper with convolutions.” In Computer Vision and Pattern Recognition, pages 1–9, 2015.

[5]	Danyang Sun, Tongzheng Ren, Chongxun Li, Hang Su, and Jun Zhu. “Learning to write stylized Chinese characters by reading a handful of examples.” In IJCAI, 2018.

[6]	F. A. Gers., J. Schmidhuber. and F. Cummins., “Learning to forget: Continual prediction with LSTM.” 9th International Conference on Artificial Neural Networks, Institution of Engineering and Technology, vol.5, pp. 850–855, 1999.

[7]	Ian J. Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, Yoshua Bengio(2014). “Generative Adversarial Networks.” stat.ML. arXiv:1406.2661v1.

[8]	Ishaan Gulrajani, Faruk Ahmed, Martin Arjovsky, Vincent Dumoulin, and Aaron C Courville. “Improved training of Wasserstein GANs.” In NeurIPS, 2017.

[9]	Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A Efros. “Unpaired image-to-image translation using cycle consistent adversarial networks.” In ICCV, 2017.

[10]	Jie Chang, Yujun Gu, Ya Zhang, and Yan-Feng Wang. “Chinese handwriting imitation with hierarchical generative adversarial network.” In BMVC, 2018.

[11]	M. Arjovsky, S. Chintala, and L. Bottou. “Wasserstein GAN.” arXiv preprint arXiv:1701.07875, 2017.

[12]	M. Mirza and S. Osindero. “Conditional generative adversarial nets.” Computer Science, pages 2672–2680,2014.

[13]	Melvin Johnson, Mike Schuster, Quoc V. Le, Maxim Krikun, Yonghui Wu, Zhifeng Chen and Nikhil Thorat. “Google’s Multilingual Neural Machine Translation System: Enabling Zero-Shot Translation.” Computation and Language,2017.

[14]	Olaf Ronneberger, Philipp Fischer, and Thomas Brox. “UNet: Convolutional networks for biomedical image segmentation.” In MICCAI, 2015.

[15]	Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, and Alexei A Efros. “Image-to-image translation with conditional adversarial networks.” In CVPR, 2017.

[16]	Pengyuan Lyu, Xiang Bai, Cong Yao, Zhen Zhu, Tengteng Huang, and Wenyu Liu. “Auto-encoder guided GAN for Chinese calligraphy synthesis.” In ICDAR, 2017.

[17]	Shan-Jean Wu, Chih-Yuan Yang and Jane Yung-jen Hsu. “CalliGAN: Style and structure-aware Chinese calligraphy character generator.” In CVPR, 2020.

[18]	Yuchen Tian. “zi2zi: Master Chinese calligraphy with conditional adversarial networks.” https://kaonashi-tyc. github.io/2017/04/06/zi2zi.html, 2017.

[19]	Yaniv Taigman, Adam Polyak, and Lior Wolf. “Unsupervised cross-domain image generation.” arXiv preprint arXiv:1611.02200, 2016.

[20]	Yexun Zhang, Ya Zhang, and Wenbin Cai. “Separating style and content for generalized style transfer.” In CVPR, 2018.

[21]	Yue Jiang, Zhouhui Lian, Yingmin Tang, and Jianguo Xiao. “DCFont: an end-to-end deep Chinese font generation system.” In SIGGRAPH Asia. 2017.

[22]	Yunjey Choi, Minje Choi, Munyoung Kim, Jung-Woo Ha, Sunghun Kim, and Jaegul Choo. “StarGAN: Unified generative adversarial networks for multi-domain image-to-image translation.” In CVPR, 2018.

[23]	Zhou Wang, Alan C. Bovik, Hamid R. Sheikh, and Eero P. Simoncelli. “Image quality assessment: from error visibility to structural similarity.” TIP, 13(4):600–612, 2004.