Research on Deformed Character Generation Method based on CGAN

Tiantian Zhao; Linghua Guo; Xiaofeng Zhao; Fengshou Tao

doi:10.6919/ICJE.202511_11(11).0004

Authors

Tiantian Zhao
Linghua Guo
Xiaofeng Zhao
Fengshou Tao

DOI:

https://doi.org/10.6919/ICJE.202511_11(11).0004

Keywords:

Text Anti-counterfeiting; Deformed Characters;

Abstract

A deformed character generation model based on Conditional Generative Adversarial Network (CGAN) was developed to reduce the risk of tampering with contract texts. Taking advantage of the characteristics of Chinese characters with "dot" strokes, which have high independence and good deformation concealment, a dataset of "dot" stroke deformed characters was constructed based on edge detection, providing training samples for the Conditional Generative Adversarial Network (CGAN). Based on this, a "dot" stroke deformed character generation model using CGAN was established to achieve the conversion from normal characters to "dot" stroke deformed characters. The model accuracy was verified using Structural Similarity Index (SSIM) and Peak Signal-to-Noise Ratio (PSNR). The "dot" stroke deformed characters were applied to anti-tampering of contract texts. The verification results show that compared with normal characters, the "dot" stroke deformed characters generated by the model have an average SSIM of 97.86% and an average PSNR of 19.72 dB, verifying the effectiveness of the model. Compared with manually designed "dot" stroke deformed characters, the average SSIM and PSNR are further improved to 98.44% and 28.58 dB, indicating that the model has high accuracy. After recognition of the replaced "dot" stroke deformed characters in contract texts, the average SSIM reaches 97.82% and the PSNR is 22.82 dB. The "dot" stroke deformed character generation model established in this paper has high accuracy, providing theoretical and practical references for the field of text anti-counterfeiting.

Downloads

Download data is not yet available.

References

[1] ZHU Q Q. Research on Text Steganography Based on Chinese Character Stroke Fing-Tuning[D].Hangzhou Dianzi University, 2023.

[2] LIU D, CHEN S, ZHOU M T. Text Digital Watermarking Technology Based on Topology of Alphabetic Symbol[J]. Journal of Chinese Computer Systems, 2007, (05): 812-815.

[3] Chang J, Gu Y. Chinese typography transfer[J]. arXiv preprint arXiv: 1707. 04904, 2017.

[4] Isola P, Zhu J Y, Zhou T, et al. Image-to-image translation with conditional adversarial networks[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2017: 1125-1134.

[5] Xiao C, Zhang C, Zheng C. Fontcode: Embedding information in text documents using glyph perturbation[J]. ACM Transactions on Graphics (TOG), 2018, 37(2): 1-16.

[6] SUN S, ZHANG W M, FANG H, et al. Automatic generation of Chinese document watermarking fonts[J]. Journal of Image and Graphics, 2022, 27(01): 262-276.

[7] Teng Shaohua, Kong Lingrui. Chinese Font Style Transfer Based on Generative Adversarial Networks[J]. Application Research of Computers,2019,36(10):3164-3167.DOI:10.19734/j.issn.1001-3695.2018.04. 0330

[8] Bai Haijuan, Zhou Wei, Wang Cunrui, et al. Font Style Transfer Method Based on Generative Adversarial Networks[J]. Journal of Dalian Minzu University,2019,21(03):250-256.DOI:10.13744/j.cnki.cn21-1431/g4.2019.03.012.

[9] XIANG T, YANG Q H, DENG Y, et al. Combining Canny edge detection algorithm and k-means algorithm for image text recognition[J]. Neijiang Science and Technology Journal, 2020, 41(10): 39-41+61.

[10] GUO S L, PIAO Z J. Statistical Report on the Number of Strokes in the GB13000.1 Character Set: Chinese Character Order (Stroke Order) Specification[J]. Modern Chinese, 2006, (11): 39-40.