Атаки на основе вредоносных возмущений на системы обработки изображений и методы защиты от них

Системы, реализующие технологии искусственного интеллекта, получили широкое распространение благодаря их эффективности в решении прикладных задач, включая компьютерное зрение. Обработка изображений посредством нейронных сетей применяется в критически важных для безопасности системах. В то же время использование искусственного интеллекта сопряжено с характерными угрозами, к которым относится и нарушение работы моделей машинного обучения. Феномен провокации некорректного отклика нейронной сети посредством внесения визуально незаметных человеку искажений впервые описан и привлек внимание исследователей в 2013 году. Методы атак на нейронные сети на основе вредоносных возмущений непрерывно совершенствовались, были предложены способы нарушения работы нейронных сетей при обработке различных типов данных и задач целевой модели. Угрозы нарушения функционирования нейронных сетей посредством указанных атак стала значимой проблемой для систем, реализующих технологии искусственного интеллекта. Таким образом, исследования в области противодействия атакам на основе вредоносных возмущений являются весьма актуальными. В данной статье представлено описание актуальных атак, приведен обзор и сравнительный анализ таких атак на системы обработки изображений с использованием искусственного интеллекта. Сформулированы подходы к классификации атак на основе вредоносных возмущений. Рассмотрены методы защиты от подобных атак, выявлены их недостатки. Показаны ограничения применяемых методов защиты, снижающие эффективность противодействия атакам. Предложены подходы по обнаружению и устранению вредоносных возмущений.

1. Goldberg Y. A primer on neural network models for natural language processing // Journal of Artifcial Intelligence Research. 2016. V. 57. P. 345–420. https://doi.org/10.1613/jair.4992

2. Nassif A.B., Shahin I., Attili I., Azzeh M., Shaalan K. Speech recognition using deep neural networks: A systematic review // IEEE Access. 2019. V. 7. P. 19143–19165. https://doi.org/10.1109/access.2019.2896880

3. Almabdy S., Elrefaei L. Deep convolutional neural network-based approaches for face recognition // Applied Sciences. 2019. V. 9. N 20. P. 4397. https://doi.org/10.3390/app9204397

4. Khan M.Z., Harous S., Hassan S. U., Khan M. U. G., Iqbal R., Mumtaz S. Deep unifed model for face recognition based on convolution neural network and edge computing // IEEE Access. 2019. V. 7. P. 72622–72633. https://doi.org/10.1109/access.2019.2918275

5. Zhang Y., Shi D., Zhan X., Cao D., Zhu K., Li Z. Slim-ResCNN: A deep residual convolutional neural network for fngerprint liveness detection // IEEE Access. 2019. V. 7. P. 91476–91487. https://doi.org/10.1109/access.2019.2927357

6. Sarvamangala D.R., Kulkarni R.V. Convolutional neural networks in medical image understanding: a survey // Evolutionary Intelligence. 2022. V. 15. N 1. P. 1–22. https://doi.org/10.1007/s12065-020-00540-3

7. Mahmood M., Al-Khateeb B., Alwash W. A review on neural networks approach on classifying cancers // IAES International Journal of Artifcial Intelligence. 2020. V. 9. N 2. P. 317–326. http://doi.org/10.11591/ijai.v9.i2.pp317-326

8. Singh V., Singh S., Gupta P. Real-time anomaly recognition through CCTV using neural networks // Procedia Computer Science. 2020. V. 173. P. 254–263. https://doi.org/10.1016/j.procs.2020.06.030

9. Severino A., Curto S., Barberi S., Arena F., Pau G. Autonomous vehicles: an analysis both on their distinctiveness and the potential impact on urban transport systems // Applied Sciences. 2021. V. 11. N 8. P. 3604. https://doi.org/10.3390/app11083604

10. Wang L., Fan X., Chen J., Cheng J., Tan J., Ma X. 3D object detection based on sparse convolution neural network and feature fusion for autonomous driving in smart cities // Sustainable Cities and Society. 2020. V. 54. P. 102002. https://doi.org/10.1016/j.scs.2019.102002

11. Chen L., Lin S., Lu X., Cao D., Wu H., Guo C., Wang F. Y. Deep neural network based vehicle and pedestrian detection for autonomous driving: A survey // IEEE Transactions on Intelligent Transportation Systems. 2021. V. 22. N 6. P. 3234–3246. https://doi.org/10.1109/tits.2020.2993926

12. Chen P. Y., Liu S. Holistic adversarial robustness of deep learning models // Proceedings of the AAAI Conference on Artifcial Intelligence. 2023. V. 37. N 13. P. 15411–15420. https://doi.org/10.1609/aaai.v37i13.26797

13. Huang X., Kroening D., Ruan W., Sharp J., Sun Y., Thamo E., Min W., Yi X. A survey of safety and trustworthiness of deep neural networks: Verifcation, testing, adversarial attack and defence, and interpretability // Computer Science Review. 2020. V. 37. P. 100270. https://doi.org/10.1016/j.cosrev.2020.100270

14. Szegedy C., Zaremba W., Sutskever I., Bruna J., Erhan D., Goodfellow I., Fergus R. Intriguing properties of neural networks // arXiv. 2013. arXiv:1312.6199. https://doi.org/10.48550/arXiv.1312.6199

15. Song Y., Shu R., Kushman N., Ermon S. Constructing unrestricted adversarial examples with generative models // Advances in Neural Information Processing Systems. 2018. V. 31.

16. Sayghe A., Zhao J., Konstantinou C. Evasion attacks with adversarial deep learning against power system state estimation // Proc. of the 2020 IEEE Power & Energy Society General Meeting (PESGM). 2020. P. 1–5. https://doi.org/10.1109/pesgm41954.2020.9281719

17. Goodfellow I.J., Shlens J., Szegedy C. Explaining and harnessing adversarial examples // arXiv. 2014. arXiv:1412.6572. https://doi.org/10.48550/arXiv.1412.6572

18. Paul R., Schabath M., Gillies R., Hall L., Goldgof D. Mitigating adversarial attacks on medical image understanding systems // Proc. of the 2020 IEEE 17th International Symposium on Biomedical Imaging (ISBI). 2020. P. 1517–1521. https://doi.org/10.1109/isbi45749.2020.9098740

19. Rozsa A., Rudd E.M., Boult T.E. Adversarial diversity and hard positive generation // Proc. of the IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). 2016. P. 25– 32. https://doi.org/10.1109/cvprw.2016.58

20. Dong Y., Liao F., Pang T., Su H., Zhu J., Hu X., Li J. Boosting adversarial attacks with momentum // Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2018. P. 9185–9193. https://doi.org/10.1109/cvpr.2018.00957

21. Miyato T., Maeda S.I., Koyama M., Ishii S. Virtual adversarial training: a regularization method for supervised and semi-supervised learning // IEEE Transactions on Pattern Analysis and Machine Intelligence. 2019. V. 41. N 8. P. 1979–1993. https://doi.org/10.1109/tpami.2018.2858821

22. Kurakin A., Goodfellow I.J., Bengio S. Adversarial examples in the physical world // Artifcial Intelligence Safety and Security. Chapman and Hall/CRC, 2018. P. 99–112. https://doi.org/10.1201/9781351251389-8

23. Mądry A., Makelov A., Schmidt L., Tsipras D., Vladu A. Towards deep learning models resistant to adversarial attacks // Stat. 2017. V. 1050. P. 9.

24. Xie C., Zhang Z., Zhou Y., Bai S., Wang J., Ren Z., Yuille A.L. Improving transferability of adversarial examples with input diversity // Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2019. P. 2730–2739. https://doi.org/10.1109/cvpr.2019.00284

25. Dong X., Han J., Chen D., Liu J., Bian H., Ma Z., Li H., Wang X., Zhang W., Yu N. Robust superpixel-guided attentional adversarial attack // Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2020. P. 12895–12904. https://doi.org/10.1109/cvpr42600.2020.01291

26. Sriramanan G., Addepalli S., Baburaj A. Guided adversarial attack for evaluating and enhancing adversarial defenses // Advances in Neural Information Processing Systems. 2020. V. 33. P. 20297–20308.

27. Rony J., Hafemann L.G., Oliveira L.S., Ayed I.B., Sabourin R., Granger E. Decoupling direction and norm for effcient gradientbased L2 adversarial attacks and defenses // Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2019. P. 4322–4330. https://doi.org/10.1109/cvpr.2019.00445

28. Moosavi-Dezfooli S.M., Fawzi A., Frossard P. DeepFool: a simple and accurate method to fool deep neural networks // Proc. of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2016. P. 2574–2582. https://doi.org/10.1109/cvpr.2016.282

29. Carlini N., Wagner D. Towards evaluating the robustness of neural networks // Proc. of the IEEE Symposium on Security and Privacy (SP). 2017. P. 39–57. https://doi.org/10.1109/sp.2017.49

30. Yao Z., Gholami A., Xu P., Keutzer K., Mahoney M. W. Trust region based adversarial attack on neural networks // Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2019. P. 11350–11359. https://doi.org/10.1109/cvpr.2019.01161

31. Papernot N., McDaniel P., Jha S., Fredrikson M., Celik Z. B., Swami A. The limitations of deep learning in adversarial settings // Proc. of the 2016 IEEE European Symposium on Security and Privacy (EuroS&P). 2016. P. 372–387. https://doi.org/10.1109/eurosp.2016.36

32. Su J., Vargas D.V., Sakurai K. One pixel attack for fooling deep neural networks // IEEE Transactions on Evolutionary Computation. 2019. V. 23. N 5. P. 828–841. https://doi.org/10.1109/tevc.2019.2890858

33. Moosavi-Dezfooli S.M., Fawzi A., Fawzi O., Frossard P. Universal adversarial perturbations // Proc. of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2017. P. 1765– 1773. https://doi.org/10.1109/cvpr.2017.17

34. Brendel W., Rauber J., Bethge M. Decision-based adversarial attacks: Reliable attacks against black-box machine learning models // Advances in Reliably Evaluating and Improving Adversarial Robustness. 2021. P. 77.

35. Chen J., Jordan M.I., Wainwright M.J. HopSkipJumpAttack: A queryeffcient decision-based attack // Proc. of the 2020 IEEE Symposium on Security and Privacy (SP). 2020. P. 1277–1294. https://doi.org/10.1109/sp40000.2020.00045

36. Liu Y., Moosavi-Dezfooli S.M., Frossard P. A geometry-inspired decision-based attack // Proc. of the IEEE/CVF International Conference on Computer Vision (ICCV). 2019. P. 4890–4898. https://doi.org/10.1109/iccv.2019.00499

37. Rahmati A., Moosavi-Dezfooli, S.M., Frossard P., Dai H. GeoDA: a geometric framework for black-box adversarial attacks // Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2020. P. 8446–8455. https://doi.org/10.1109/cvpr42600.2020.00847

38. Du J., Zhang H., Zhou J.T., Yang Y., Feng J. Query-effcient meta attack to deep neural networks // Proc. of the International Conference on Learning Representations. 2020.

39. Li J., Ji R., Liu H., Liu J., Zhong B., Deng C., Tian Q. Projection & probability-driven black-box attack // Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2020. P. 362–371. https://doi.org/10.1109/cvpr42600.2020.00044

40. Li H., Xu X., Zhang X., Yang S., Li B. QEBA: Query-effcient boundary-based blackbox attack // Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2020. P. 1221– 1230. https://doi.org/10.1109/cvpr42600.2020.00130

41. Cheng M., Singh S., Chen P., Chen P.Y., Liu S., Hsieh C.J. Sign-OPT: A query-effcient hard-label adversarial attack // Proc. of the International Conference on Learning Representations. 2020.

42. Brunner T., Diehl F., Le M.T., Knoll A. Guessing smart: Biased sampling for effcient black-box adversarial attacks // Proc. of the IEEE/CVF International Conference on Computer Vision (ICCV). 2019. P. 4958–4966. https://doi.org/10.1109/iccv.2019.00506

43. Maho T., Furon T., Le Merrer E. SurFree: a fast surrogate-free blackbox attack // Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2021. P. 10430–10439. https://doi.org/10.1109/cvpr46437.2021.01029

44. Shi Y., Han Y., Tian Q. Polishing decision-based adversarial noise with a customized sampling // Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2020. P. 1030– 1038. https://doi.org/10.1109/cvpr42600.2020.00111

45. Huang Z., Zhang T. Black-box adversarial attack with transferable model-based embedding // Proc. of the International Conference on Learning Representations. 2020.

46. Zhou M., Wu J., Liu Y., Liu S., Zhu C. DaST: Data-free substitute training for adversarial attacks // Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2020. P. 234– 243. https://doi.org/10.1109/cvpr42600.2020.00031

47. Zou J., Pan Z., Qiu J., Liu X., Rui T., Li W. Improving the transferability of adversarial examples with resized-diverse-inputs, diversity-ensemble and region ftting // Lecture Notes in Computer Science. 2020. V. 12367. P. 563–579. https://doi.org/10.1007/978-3-030-58542-6_34

48. Wang X., He K. Enhancing the transferability of adversarial attacks through variance tuning // Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2021. P. 1924–1933. https://doi.org/10.1109/cvpr46437.2021.00196

49. Wu W., Su Y., Lyu M.R., King I. Improving the transferability of adversarial samples with adversarial transformations // Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2021. P. 9024–9033. https://doi.org/10.1109/cvpr46437.2021.00891

50. Hosseini H., Poovendran R. Semantic adversarial examples // Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). 2018. P. 1614–1619. https://doi.org/10.1109/cvprw.2018.00212

51. Engstrom L., Tran B., Tsipras D., Schmidt L., Madry A. A rotation and a translation suffce: Fooling cnns with simple transformations [Электронный ресурс]. URL: https://openreview.net/forum?id=BJfvknCqFQ (дата обращения: 29.05.2023).

52. Joshi A., Mukherjee A., Sarkar S., Hegde C. Semantic adversarial attacks: Parametric transformations that fool deep classifers // Proc. of the IEEE/CVF International Conference on Computer Vision (ICCV). 2019. P. 4773–4783. https://doi.org/10.1109/iccv.2019.00487

53. Liu A., Wang J., Liu X., Cao B., Zhang C., Yu H. Bias-based universal adversarial patch attack for automatic check-out // Lecture Notes in Computer Science. 2020. V. 12358. P. 395–410. https://doi.org/10.1007/978-3-030-58601-0_24

54. Swathi P., Sk S. DeepFake creation and detection: A survey // Proc. of the 2021 Third International Conference on Inventive Research in Computing Applications (ICIRCA). 2021. P. 584–588. https://doi.org/10.1109/icirca51532.2021.9544522

55. Chadha A., Kumar V., Kashyap S., Gupta M. Deepfake: An Overview // Lecture Notes in Networks and Systems. 2021. V. 203. P. 557–566. https://doi.org/10.1007/978-981-16-0733-2_39

56. Nakka K.K., Salzmann M. Indirect local attacks for context-aware semantic segmentation networks // Lecture Notes in Computer Science. 2020. V. 12350. P. 611–628. https://doi.org/10.1007/978-3-030-58558-7_36

57. He Y., Rahimian S., Schiele B., Fritz M. Segmentations-leak: Membership inference attacks and defenses in semantic image segmentation // Lecture Notes in Computer Science. 2020. V. 12368. P. 519–535. https://doi.org/10.1007/978-3-030-58592-1_31

58. Choi J.H. Zhang H., Kim J.H., Hsieh C.J., Lee J.S. Evaluating robustness of deep image super-resolution against adversarial attacks // Proc. of the IEEE/CVF International Conference on Computer Vision (ICCV). 2019. P. 303–311. https://doi.org/10.1109/iccv.2019.00039

59. Jiang L., Ma X., Chen S., Bailey J., Jiang Y.G. Black-box adversarial attacks on video recognition models // Proc. of the 27th ACM International Conference on Multimedia. 2019. P. 864–872. https://doi.org/10.1145/3343031.3351088

60. Li S., Aich A., Zhu S., Asif S., Song C., Roy-Chowdhury A., Krishnamurthy S. Adversarial attacks on black box video classifers: Leveraging the power of geometric transformations // Advances in Neural Information Processing Systems. 2021. V. 34. P. 2085–2096.

61. Chen X., Li S., Huang H. Adversarial attack and defense on deep neural network-based voice processing systems: An overview // Applied Sciences. 2021. V. 11. N 18. P. 8450. https://doi.org/10.3390/app11188450

62. Kwon H., Kim Y., Yoon H., Choi D. Selective audio adversarial example in evasion attack on speech recognition system // IEEE Transactions on Information Forensics and Security. 2020. V. 15. P. 526–538. https://doi.org/10.1109/tifs.2019.2925452

63. Usama M., Qayyum A., Qadir J., Al-Fuqaha A. Black-box adversarial machine learning attack on network traffc classifcation // Proc. of the 15th International Wireless Communications & Mobile Computing Conference (IWCMC). 2019. P. 84–89. https://doi.org/10.1109/iwcmc.2019.8766505

64. Imam N.H., Vassilakis V.G. A survey of attacks against twitter spam detectors in an adversarial environment // Robotics. 2019. V. 8. N 3. P. 50. https://doi.org/10.3390/robotics8030050

65. Zhong H., Liao C., Squicciarini A.C., Zhu S., Miller D. Backdoor embedding in convolutional neural network models via invisible perturbation // Proc. of the Tenth ACM Conference on Data and Application Security and Privacy. 2020. P. 97–108. https://doi.org/10.1145/3374664.3375751

66. Liu X., Yang H., Liu Z., Song L., Li H., Chen Y. Dpatch: An adversarial patch attack on object detectors // arXiv. 2018. arXiv:1806.02299. https://doi.org/10.48550/arXiv.1806.02299

67. Liu Y., Ma X., Bailey J., Lu F. Refection backdoor: A natural backdoor attack on deep neural networks // Lecture Notes in Computer Science. 2020. V. 12355. P. 182–199. https://doi.org/10.1007/978-3-030-58607-2_11

68. Nguyen A., Tran A. WaNet - imperceptible warping-based backdoor attack // Proc. of the International Conference on Learning Representations. 2021.

69. Костюмов В.В. Обзор и систематизация атак уклонением на модели компьютерного зрения // International Journal of Open Information Technologies. 2022. Т. 10. № 10. С. 11–20.

70. Papernot N., McDaniel P., Wu X., Jha S., Swami A. Distillation as a defense to adversarial perturbations against deep neural networks // Proc. of the 2016 IEEE Symposium on Security and Privacy (SP). 2016. P. 582–597. https://doi.org/10.1109/sp.2016.41

71. Steihaug T. The conjugate gradient method and trust regions in large scale optimization // SIAM Journal on Numerical Analysis. 1983. V. 20. N 3. P. 626–637. https://doi.org/10.1137/0720042

72. Curtis A.R., Powell M.J.D., Reid J.K. On the estimation of sparse Jacobian matrices // IMA Journal of Applied Mathematics. 1974. V. 13. N 1. P. 117–120. https://doi.org/10.1093/imamat/13.1.117

73. Niebur E. Saliency map // Scholarpedia. 2007. V. 2. N 8. С. 2675. https://doi.org/10.4249/scholarpedia.2675

74. Das S., Suganthan P.N. Differential evolution: A survey of the stateof-the-art // IEEE Transactions on Evolutionary Computation. 2011. V. 15. N 1. P. 4–31. https://doi.org/10.1109/tevc.2010.2059031

75. Badrinarayanan V., Kendall A., Cipolla R. SegNet: A deep convolutional encoder-decoder architecture for image segmentation // IEEE Transactions on Pattern Analysis and Machine Intelligence. 2017. V. 39. N 12. P. 2481–2495. https://doi.org/10.1109/tpami.2016.2644615

76. Lowd D., Meek C. Adversarial learning // Proc. of the Eleventh ACM SIGKDD International Conference on Knowledge Discovery in Data Mining. 2005. P. 641–647. https://doi.org/10.1145/1081870.1081950

77. Xu W., Evans D., Qi Y. Feature squeezing: Detecting adversarial examples in deep neural networks // Proc. of the 2018 Network and Distributed System Security Symposium (NDSS). 2018.

78. Liao F., Liang M., Dong Y., Pang T., Hu X., Zhu J. Defense against adversarial attacks using high-level representation guided denoiser // Proc. of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2018. P. 1778–1787. https://doi.org/10.1109/cvpr.2018.00191

79. Zhang D., Ye M., Gong C., Zhu Z., Liu Q. Black-box certifcation with randomized smoothing: A functional optimization based framework // Advances in Neural Information Processing Systems. 2020. V. 33. P. 2316–2326.

80. Fischer M., Baader M., Vechev M. Certifed defense to image transformations via randomized smoothing // Advances in Neural Information Processing Systems. 2020. V. 33. P. 8404–8417.

81. Yang R., Chen X.Q., Cao T.J. APE-GAN++: An improved APE-GAN to eliminate adversarial perturbations // IAENG International Journal of Computer Science. 2021. V. 48. N 3. P. 827–844.

82. Glenn T.C., Zare A., Gader P.D. Bayesian fuzzy clustering // IEEE Transactions on Fuzzy Systems. 2015. V. 23. N 5. P. 1545–1561. https://doi.org/10.1109/tfuzz.2014.2370676

83. Plackett R.L. Karl Pearson and the chi-squared test // International Statistical Review / Revue Internationale de Statistique. 1983. V. 51. N 1. P. 59–72. https://doi.org/10.2307/1402731

84. McLachlan G.J. Mahalanobis distance // Resonance. 1999. V. 4. N 6. P. 20–26. https://doi.org/10.1007/BF02834632