To address the limited transferability of existing adversarial face detection methods across datasets, a novel method based on the inconsistency of sensitivity to variable image resolution is proposed. The sensitivity changes during image transformations, such as super-resolution and denoising, are analyzed, and cosine similarity is employed to quantify the differences in facial features before and after transformation, enabling the detection of adversarial face attacks. Experiments on datasets such as Labeled Faces in the Wild (LFW) and CelebA-HQ have demonstrated that a detection accuracy rate of over 90% under six types of adversarial attacks is achieved by the proposed method.
为解决现有对抗人脸检测方法跨数据集泛化能力弱的问题,受变分辨率的方法和基于图像变换的对抗检测方法的启发,提出了一种基于图像变分辨率敏感度不一致性的对抗性人脸检测方法,其核心思想是利用对抗性人脸图像在变换不同分辨率后,对超分、去噪等图像变换操作的敏感度差异来检测攻击。在公开的经典人脸数据库(Labeled Faces in the Wild, LFW)[26]和CelebA-HQ[27]等数据集上进行了多种攻击检测的实验,结果表明,在6种对抗攻击条件下,检测准确率均达到90%以上,从而验证了该方法的有效性。
本文方法采用计算效率高、适用性强的增强深度残差网络(Enhanced Deep Residual Networks, EDSR)模型[29]进行微调,保留了原有的网络结构,具体结构如图2所示。在该模块中,预期超分网络如实表达图像高频信息的细节,因此仅修改了网络的输入输出层,以适应数据集图像大小,从而更好地验证在不同图像尺度下的效果。
DUH L, SHIH, ZENGD, et al. The elements of end-to-end deep face recognition: a survey of recent advances[J]. ACM Computing Surveys, 2020,54():No.212.
[2]
TAIGMANY, YANGM, RANZATOM, et al. Deepface: closing the gap to human-level performance in face verification[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway, USA: IEEE, 2014:1701-1708.
[3]
SZEGEDYC, ZAREMBAW, SUTSKEVERI, et al. Intriguing properties of neural networks[DB/OL]. (2013-12-05)[2024-04-03].
[4]
GOODFELLOWI J, SHLENSJ, SZEGEDYC. Explaining and harnessing adversarial examples[DB/OL]. (2014-12-20)[2024-04-03].
[5]
KURAKINA, GOODFELLOWI J, BENGIOS. Adversarial machine learning at scale[DB/OL]. (2016-11-04)[2024-04-03].
[6]
MADRYA, MAKELOVA, SCHMIDTL, et al. Towards deep learning models resistant to adversarial attacks[DB/OL]. (2017-06-19)[2024-04-03].
[7]
TRAMERF, KURAKINA, PAPERNOTN, et al. Ensemble adversarial training: attacks and defenses[DB/OL].(2017-05-19)[2024-04-03].
[8]
DONGY P, LIAOF Z, PANGT Y, et al. Boosting adversarial attacks with momentum[C]∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway, USA: IEEE, 2018:9185-9193.
[9]
WUW B, SUY X, LYUM R, et al. Improving the transferability of adversarial samples with adversarial transformations[C]∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway, USA: IEEE, 2021:9020-9029.
DEBD, ZHANGJ, JAINA K. AdvFaces: adversarial face synthesis[C]∥Proceedings of the IEEE International Joint Conference on Biometrics. Piscataway, USA: IEEE, 2020. DOI:10.1109/IJCB48548.2020.9304898 .
[12]
QIUH V, XIAOC W, YANGL, et al. SemanticAdv: generating adversarial examples via attribute-conditional image editing [DB/OL]. (2019-06-19)[2024-04-03].
[13]
JIAS, YINB L, YAOT P, et al. Adv-Attribute: inconspicuous and transferable adversarial attack on face recognition [DB/OL]. (2022-10-13)[2024-06-19].
[14]
LIQ, HUY X, LIUY, et al. Discrete point-wise attack is not enough: generalized manifold adversarial attack for face recognition[C]∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway, USA: IEEE, 2023:20575-20584.
[15]
MASSOLIF V, CARRARAF, AMATOG, et al. Detection of face recognition adversarial attacks[DB/OL].(2019-12-09)[2024-04-03].
[16]
AGARWALA, SINGHR, VATSAM, et al. Are image-agnostic universal adversarial perturbations for face recognition difficult to detect?[C]∥Proceedings of the IEEE 9th International Conference on Biometrics Theory, Applications and Systems. Piscataway, USA:IEEE,2018. DOI:10.1109/BATS.2018.8698548 .
[17]
TAOG H, MAS Q, LIUY Q, et al. Attacks meet interpretability: attribute-steered detection of adversarial samples [DB/OL]. (2018-10-27)[2024-04-03].
[18]
DENGZ J, YANGX, XUS Z, et al. LiBRe: a practical bayesian approach to adversarial detection[C]∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway,USA: IEEE, 2021:972-982.
[19]
DEBD, LIUX, JAINA K. FaceGuard: a self-supervised defense against adversarial face images[C]∥Proceedings of the IEEE 17th International Conference on Automatic Face and Gesture Recognition. Piscataway, USA: IEEE, 2023. DOI:10.1109/FG57933.2023.10042617 .
[20]
LINGH F, SUNF R, ZHANGJ Y, et al. General adversarial perturbation simulating:protect unknown system by detecting unknown adversarial faces[C]∥Proceedings of the International Joint Conference on Neural Networks. Piscataway, USA:IEEE,2023. DOI:10.1109/IJCNN5316 3.2023.10374958 .
[21]
WANGQ, XIANY Q, LINGH F, et al. Detecting adversarial faces using only real face self-perturbations[C]∥Proceedings of the Thirty-Second International Joint Conference on Artificial Intelligence. California,USA: International Joint Conferences on Artificial Intelligence Organization, 2023:1488-1496.
[22]
SCHROFFF, KALENICHENKOD, PHILBINJ. Facenet: a unified embedding for face recognition and clustering[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway, USA: IEEE, 2015:815-823.
[23]
AGARWALA, SINGHR, VATSAM, et al. Image transformation-based defense against adversarial perturbation on deep learning models[J]. IEEE Transactions on Dependable and Secure Computing, 2021,18(4):2106-2121.
[24]
TIANJ Y, ZHOUJ T, LIY M, et al. Detecting adversarial examples from sensitivity inconsistency of spatial-transform domain[C]∥Proceedings of the AAAI Conference on Artificial Intelligence. Washington, USA: AAAI Press, 2021:9877-9885.
[25]
MUSTAFAA, KHANS H, HAYATM, et al. Image super-resolution as a defense against adversarial attacks[J]. IEEE Transactions on Image Processing, 2019,29:1711-1724.
[26]
FENWICKS R, KOSSAIFIJ, SANDBERGD, et al. Labeled faces in the wild: updates and new reporting methods [R]. Massachusetts,USA: University of Massachusetts, 2014:1-5.
[27]
LIUZ W, LUOP, WANGX G, et al. Deep learning face attributes in the wild[C]∥Proceedings of the IEEE International Conference on Computer Vision. Piscataway, USA: IEEE, 2015:3730-3738.
LIM B, SON S, KIMH, et al. Enhanced deep residual networks for single image super-resolution[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops. Piscataway,USA:IEEE,2017:1132-1140.
DENGJ K, GUOJ, XUEN, et al. ArcFace: additive angular margin loss for deep face recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022, 44(10): 5962-5979.
[32]
KIMH. Torchattacks: a PyTorch repository for adversarial attacks[DB/OL]. (2020-09-24)[2024-04-03].