曾辉/(Zeng Hui）
计算机科学与技术学院教师 副研究员 硕士生导师
电子邮件:zengh5@mail2.sysu.edu.cn
办公地址: 东九2-68
研究方向:博弈论, 图像取证, 对抗环境下机器学习
社会兼职:中国图象图形学学会 数字媒体取证与安全专委会 委员
主讲课程:多媒体安全，数字取证技术实验，信息安全新技术专题等
个人简介/最新近况
欢迎对媒体安全，视觉模型安全等方向有兴趣的中外老师和同学前来交流。
学术主页：https://scholar.google.com/citations?user=__LlM6MAAAAJ
citation: 400+, H-index: 11
代码仓库: https://www.github.com/zengh5
招生信息: 每年招收1-3名国内或国际硕士研究生，申请者首次联系时需至少阅读课题组研究领域内论文一篇。
论文、著作、专利等
论文：
1. 视频/图像取证
[1] Zeng, Hui and Kang, Xiangui, “Fast source camera identification using content adaptive guided image filter”, Journal of forensic sciences, 61(2): 520–526, 2016
Code: https://github.com/zengh5/SCI_CAGF
[2] Zeng, H. and Peng, A. and Lin, X. and Luo, S., “Source smartphone identification for digital zoomed images”, Proceedings of the ACM Turing Celebration Conference-China, pp. 1–6, 2019.
[3] Zeng, Hui and Wan, Y. and Deng, K. and Peng, A., “Source Camera Identification with Dual-Tree Complex Wavelet Transform” IEEE Access, 8: 18874–18883. Code: https://github.com/zengh5/SCI_DTCWT
Paper: https://ieeexplore.ieee.org/document/8966247
[4] Zeng, Hui and Zhan, Yifeng and Kang, Xiangui and Lin, Xiaodan, “Image splicing localization using PCA-based noise level estimation”, Multimedia Tools and Applications, 76(4): 4783–4799, 2017
Code: https://github.com/MKLab-ITI/image-forensics/tree/master/matlab_toolbox
Paper: https://link.springer.com/article/10.1007/s11042-016-3712-8
[5] Zeng, Hui, Peng, Anjie, and Lin, Xiaodan, “Exposing Image splicing with inconsistent sensor noise levels”, Multimedia Tools and Applications, 2020, 79: 26139–26154
Code: https://github.com/zengh5/Exposing-splicing-sensor-noise
Paper: https://link.springer.com/article/10.1007/s11042-020-09280-z
[6] MDM Hosseini, M. Goljan, and Hui Zeng, “Semi-Blind Image Resampling Factor Estimation for PRNU Computation,” in Proc. IS&T Int’l. Symp. on Electronic Imaging: Media Watermarking, Security, and Forensics, 2020, pp 77-1–77-11, https://doi.org/10.2352/ISSN.2470-1173.2020.4.MWSF-077
[7] Hui Zeng, MDM Hosseini, and Miroslav Goljan, “Replacing DWT with DTCWT in blind image rotation angle estimation,” in Proc. IS&T Int’l. Symp. on Electronic Imaging: Media Watermarking, Security, and Forensics, 2021, pp 275-1–275-7, https://doi.org/10.2352/ISSN.2470-1173.2021.4.MWSF-275
[8] Kun Yu, Rongsong Yang, Hui Zeng, and Anjie Peng, 'Joint estimation of image rotation angle and scaling factor', 2021 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC), pp. 1716–1721.
[9] Kun Yu, MDM. Hosseini, Anjie Peng, Hui Zeng, M. Goljan, 'Make your enemy your friend: improving image rotation angle estimation with harmonics,' ICASSP2023.
Code: https://github.com/zengh5/Rotation_angle_estimation_harmonic
Paper: https://ieeexplore.ieee.org/document/10095317/
[10] Hui Zeng, Kang Deng, Anjie Peng, ISO Setting Estimation Based on Convolutional Neural Network and Its Application in Image Forensics, IWDW2020.
Code: https://github.com/zengh5/ISONet
[11] Tong Zhang, Anjie Peng, Hui Zeng, “Ignored Details in Eyes: Exposing GAN-generated Faces by Sclera,” 2023ICONIP, pp 563–574.
Paper: https://link.springer.com/chapter/10.1007/978-981-99-8073-4_43
Code: https://github.com/10961020/Deepfake-detector-based-on-blood-vessels
2 反取证及其对策
[12] Zeng, Hui and Chen, J. and Kang, X. and Zeng, W., “Removing camera fingerprint to disguise photograph source”, 2015 IEEE International Conference on Image Processing (ICIP), 1687–1691
[13] Wu, J. and Wang, Z. and Zeng, Hui and Kang, X., “Multiple-Operation Image Anti-Forensics with WGAN-GP Framework”, 2019 APSIPA ASC, pp. 1303–1307.
[14] Zeng, Hui, Peng, Anjie, and Kang, Xiangui, 'Hiding traces of camera anonymization by Poisson blending“ Accepted by ICAIS2020.
[15] Zeng, Hui and Kang, Xiangui and Huang, Jiwu, “Mixed-strategy Nash equilibrium in the camera source identification game”, 2013ICIP, 4472–4476
[16] Zeng, Hui and Jiang, Yunwen and Kang, Xiangui and Liu, Li, “Game theoretic analysis of camera source identification”, 2013 APSIPA ASC, 1–9.
[17] Zeng, Hui and Kang, Xiangui, “Camera source identification game with incomplete information”, International Workshop on Digital Watermarking (IWDW), 192–204, 2013, (Best student paper)
[18] Zeng, Hui and Liu, J. and Yu, J. and Kang, X. and Shi, Y. and Wang, Z Jane “A framework of camera source identification Bayesian game”, IEEE transactions on cybernetics, 47(7): 1757–1768, 2016
Paper: https://ieeexplore.ieee.org/document/7469854/
[19] X. Kang, T. Qin and H. Zeng, 'Countering median filtering anti-forensics and performance evaluation of forensics against intentional attacks,' 2015 IEEE China Summit and International Conference on Signal and Information Processing (ChinaSIP). Paper: https://ieeexplore.ieee.org/document/7230449
[20] Zeng, Hui and Qin, Tengfei and Kang, Xiangui and Liu, Li, “Countering anti-forensics of median filtering”, 2014 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2704–2708
Paper: https://ieeexplore.ieee.org/document/6854091
[21] Jiang, Y. and Zeng, Hui and Kang, X. and Liu, L., “The game of countering JPEG anti-forensics based on the noise level estimation”, 2013 APSIPA ASC, 1–9.
[22] Zeng, Hui and Yu, J. and Kang, X. and Lyu, S., “Countering JPEG anti-forensics based on noise level estimation”, Science China Information Sciences, 61(3): 032103, 2018
[23] Zeng, Hui and Kang, Xiangui and Peng, Anjie, “A multi-purpose countermeasure against image anti-forensics using autoregressive model”, Neurocomputing, 189: 117–122, 2016
3 对抗样本生成/检测
[24] A. Peng, K. Deng, J. Zhang, S. Luo, H. Zeng, W. Yu, “Gradient-based adversarial image forensics,” the 27th International Conference on Neural Information Processing, pp. 417–428, 2020
[25] Kang Deng, Anjie Peng, Hui Zeng, 'Detecting C&W adversarial images based on noise addition-then-denoising,' ICIP2021, pp.3607–3611
[26] Hui Zeng, Kang Deng, Biwei Chen, Anjie Peng, 'How secure are the adversarial examples themselves?' ICASSP2022, pp. 2879–2883
Code: https://github.com/zengh5/adversarial-example-security
Paper: https://ieeexplore.ieee.org/document/9747206
[27] Zhi Lin, Anjie Peng, Hui Zeng, et al. 'Boosting transferability of adversarial example via an enhanced EULER's method,' ICASSP2023.
Paper: https://ieeexplore.ieee.org/document/10096558
[28] Zhi Lin, Anjie Peng, Hui Zeng, et al. 'An enhanced neuron attribution-based attack via pixel dropping,' ICIP2023, pp. 3439–3443.
Paper: https://ieeexplore.ieee.org/document/10222034
[29] Hui Zeng, Tong Zhang, Biwei Chen, Anjie Peng. 'Enhancing targeted transferability via suppressing high-confidence labels,' ICIP2023, pp. 3309–3313.
Paper: https://ieeexplore.ieee.org/document/10222841
Code: https://github.com/zengh5/Transferable_targeted_attack
[30] Hui Zeng, Biwei Chen, Kang Deng, Anjie Peng. 'Adversarial example detection Bayesian game,' ICIP2023, pp. 1710–1714.
Paper: https://ieeexplore.ieee.org/document/10222129
Code: github.com/zengh5/AED_BGame.
[31] H. Zeng, B. Chen, R. Yang, C. Li, A. Peng, “Towards undetectable adversarial examples: a steganographic perspective,” 2023ICONIP, pp. 172–183.
Paper: https://link.springer.com/chapter/10.1007/978-981-99-8070-3_14
Code: github.com/zengh5/Undetectable-attack
[32] C. Li, A. Peng, Hui Zeng, et al., “Fooling Downstream Classifiers via Attacking Contrastive Learning Pre-trained Models” 2023ICONIP.
[33] G. Shi, A. Peng, Hui Zeng, et al., “Neuron Attribution-Based Attacks Fooling Object Detectors” 2023ICONIP.
[34] A. Peng, K. Deng, Hui Zeng*, et al. “Detecting Adversarial Examples via Classification Difference of a Robust Surrogate Mode” 2023ICONIP.
[35] G. Shi, Z. Lin, A. Peng, Hui Zeng, “An Enhanced Transferable Adversarial Attack Against Object Detection” 2023IJCNN.
[36] H. Zeng, B. Chen, A. Peng, “Enhancing targeted transferability via feature space fine-tuning” ICASSP2024, pp. 4475–4479.
Paper: https://arxiv.org/abs/2401.02727 Code: github.com/zengh5/TA_feature_FT
[37] A. Peng; G. Shi; Z. Lin; H. Zeng; X. Yang, 'Approximating High-order Adversarial Attacks Using Runge-Kutta Methods,' 2024, Tsinghua Science and Technology.
[38] Qiang Wan, Biwei Chen, Anjie Peng, Hui Zeng, 'A whale falls, all thrive: Mitigating attention gap to improve adversarial transferability,' 2024ICPR, pp. 346–359. Code: https://github.com/britney-code/EIT-attack
[39] Hui Zeng, Sanshuai Cui, Biwei Chen, Anjie Peng, 'Everywhere Attack: Attacking Locally and Globally to Boost Targeted Transferability,' to appear in 2025AAAI.
Code: https://github.com/zengh5/Everywhere_Attack
项目
[1] 对抗样本安全性提升方法研究，2024.05-2025.05，广东省信息安全重点实验室开放课题，2023B1212060026，2万元，主持；
[2] 通过隐写嵌入概率图提升对抗样本的安全性，2022.06-2023.06，广东省信息安全重点实验室开放课题，2020B1212060078，2万元，主持；
[3] 最新图像操作技术的取证和反取证研究, 2018.07-2021.06，西南科技大学科研启动金，30万元，主持；
[4] 图像来源取证安全性研究,2018.01-2020.12, 国家自然科学基金(青年), 61702429，25万元，主持；
[5] 信息取证博弈和加密域信号,2014.01-2017, 国家自然科学基金(面上)，61379155，67万元，参与。
获奖、荣誉、学术服务
[1] 在2020年阿里天池 AI security 挑战赛, 3名实验室研究生万永财、余琨、羊荣松获得赛道一(攻击)全球第2名，赛道二(防御)全球第34名。
https://tianchi.aliyun.com/competition/entrance/531812/rankingList/0
[2] In ChinaMFS2023, 硕士生张桐荣获优秀论文奖.
https://cs.swust.edu.cn/newsdetail/news-8755
教育经历
访问学者, 2019.11-2020.11, Binghamton University, SUNY, 合作导师: M. Goljan；
博士, 2012-2016, 中山大学, 通信与信息系统专业，导师：康显桂；
硕士, 2004-2007, 南京邮电大学, 通信与信息系统专业， 导师：糜正琨；
本科, 2000-2004, 南京邮电大学, 通信工程专业。