A Video Gait Privacy Protection Algorithm Based on Sparse Adversarial Attack on Silhouette

doi:10.3969/j.issn.1671-1122.2024.01.005

Abstract

Abstract:

Deep network models can obtain human gait biometrics from video gait sequences and recognize character identities through feature matching, which threatens human privacy. Privacy protection treatments such as blurring and deformation of the human body in video images can to some extent change the appearance of the human body. Still, it is difficult to change the walking posture of the characters and cannot avoid recognition by deep network models. Moreover, this treatment often accompanies serious damage to video quality, reducing the visual usability of the video. In response to this issue, this article proposed a video gait privacy protection algorithm based on sparse adversarial attack on silhouette, which calculates effective modification positions around human silhouette in the image through adversarial attacks on gait recognition models. Compared with traditional methods, this algorithm reduces the modification of images while maintaining the same privacy protection capabilities. The optimal balance between privacy security and visual availability was obtained. The algorithm is tested on four gait recognition models using the public gait datasets CASIA-B and OUMVLP, and previous gait privacy protection methods are implemented and compared, verifying the effectiveness and availability of this algorithm in gait privacy protection.

Key words: gait privacy protection, gait recognition, sparse adversarial attack on silhouette, adversarial samples

CLC Number:

TP309

XU Ke, LI Jiayi, JIANG Xinghao, SUN Tanfeng. A Video Gait Privacy Protection Algorithm Based on Sparse Adversarial Attack on Silhouette[J]. Netinfo Security, 2024, 24(1): 48-59.

Figures/Tables 11

References 32

[1]	CHEN Xingyuan, GAO Yuanzhao, TANG Huilin, et al. Research Progress on Big Data Security Technology[J]. SCIENTIA SINICA Informationis, 2020, 50(1): 25-66. doi: 10.1360/N112019-00077 URL
	陈性元, 高元照, 唐慧林, 等. 大数据安全技术研究进展[J]. 中国科学:信息科学, 2020, 50(1):25-66.
[2]	ALIREZA S M, ALI E. Deep Gait Recognition: A Survey[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45(1): 264-284. doi: 10.1109/TPAMI.2022.3151865 URL
[3]	YU Shiqi, TAN Daoliang, TAN Tieniu. A Framework for Evaluating the Effect of View Angle, Clothing and Carrying Condition on Gait Recognition[C]// IEEE. IEEE 18th International Conference on Pattern Recognition. New York: IEEE. 2006: 441-444.
[4]	IWAMA H, OKUMURA M, MAKIHARA Y, et al. The OU-ISI Gait Database Comprising the Large Population Dataset and Performance Evaluation of Gait Recognition[J]. IEEE Transactions on Information Forensics and Security, 2012, 7(5): 1511-1521. doi: 10.1109/TIFS.2012.2204253 URL
[5]	TAKEMURA N, MAKIHARA Y, MURAMATSU D, et al. Multi-View Large Population Gait Dataset and Its Performance Evaluation for Cross-View Gait Recognition[J]. IPSJ Transactions on Computer Vision and Applications, 2018, 10(4):1-14. doi: 10.1186/s41074-017-0037-0
[6]	LI Xiang, YASUSHI M, XU Chi, et al. End-to-End Model-Based Gait Recognition Using Synchronized Multi-View Pose Constraint[C]// IEEE. IEEE/CVF International Conference on Computer Vision Workshops. New York: IEEE, 2021: 4089-4098.
[7]	LI Xiang, YASUSHI M, XU Chi, et al. End-to-End Model-Based Gait Recognition[C]// Springer. Asian Conference on Computer Vision. Heidelberg: Springer, 2020: 3-20.
[8]	ANNA S, ANTON K. Pose-Based Deep Gait Recognition[J]. IET Biometrics, 2019, 8(2): 134-143. doi: 10.1049/iet-bmt.2018.5046
[9]	TORBEN T, ALI K, JOHANNES G, et al. Gaitgraph: Graph Convolutional Network for Skeleton-Based Gait Recognition[C]// IEEE. 2021 IEEE International Conference on Image Processing (ICIP). New York: IEEE, 2021: 2314-2318.
[10]	HOU Saihui, CAO Chunshui, LIU Xu, et al. Gait Lateral Network: Learning Discriminative and Compact Representations for Gait Recognition[C]// Springer. European Conference on Computer Vision. Heidelberg: Springer, 2020: 382-398.
[11]	LIN Beibei, ZHANG Shunli, BAO Feng. Gait Recognition with Multiple-Temporal-Scale 3D Convolutional Neural Network[J]. ACM.Proceedings of the 28th ACM International Conference on Multimedia. New York: ACM, 2020: 3054-3062.
[12]	LI Xiang, YASUSHI M, XU Chi, et al. Gait Recognition via Semi-Supervised Disentangled Representation Learning to Identity and Covariate Features[C]// IEEE. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2020: 13306-13316.
[13]	CHAO Hanqing, HE Yiwei, ZHANG Junping, et al. GaitSet: Regarding Gait as a Set for Cross-View Gait Recognition[C]// ACM. Proceedings of the Thirty-Third AAAI Conference on Artificial Intelligence and Thirty-First Innovative Applications of Artificial Intelligence Conference and Ninth AAAI Symposium on Educational Advances in Artificial Intelligence. New York: ACM, 2019: 8126-8133.
[14]	ALIREZA S M, ALI E. View-Invariant Gait Recognition with Attentive Recurrent Learning of Partial Representations[J]. IEEE Transactions on Biometrics, Behavior, and Identity Science, 2021, 3(1): 124-137. doi: 10.1109/TBIOM.8423754 URL
[15]	ZHANG Ziyuan, LUAN T, LIU Feng, et al. On Learning Disentangled Representations for Gait Recognition[J]. IEEE, 2022, 44(1): 345-360.
[16]	ZHANG Jianwu, SHEN Wei, WU Zhendong. Recognition of Face Privacy Protection Using Convolutional Neural Networks[J]. Journal of Image and Graphics, 2019, 24(5): 744-752.
	章坚武, 沈炜, 吴震东. 卷积神经网络的人脸隐私保护识别[J]. 中国图象图形学报, 2019, 24(5): 744-752.
[17]	SHAWN S, EMILY W, ZHANG Jiayun, et al. Fawkes: Protecting Privacy Against Unauthorized Deep Learning Models[C]// ACM. Proceedings of the 29th USENIX Conference on Security Symposium. New York: ACM, 2020: 1589-1604.
[18]	PRACHI A, NARAYANAN P J. Person De-Identification in Videos[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2011, 21(3): 299 -310. doi: 10.1109/TCSVT.2011.2105551 URL
[19]	PARASHAR A, SHEKHAWAT R S. Protection of Gait Data Set for Preserving Its Privacy in Deep Learning Pipeline[J]. IET Biometrics, 2022(11): 557-569.
[20]	YUKI H, KAZUAKI N, NAOKO N, et al. Anonymization of Human Gait in Video Based on Silhouette Deformation and Texture Transfer[J]. IEEE Transactions on Information Forensics and Security, 2022(17): 3375-3390.
[21]	NGOC-DUNG T T, HUY H N, HOANG-QUOC, et al. An Approach for Gait Anonymization Using Deep Learning[C]// IEEE. 2017 IEEE Workshop on Information Forensics and Security (WIFS). New York: IEEE, 2017: 1-6.
[22]	NGOC-DUNG T T, HUY H N, FANG Fuming, et al. An RGB Gait Anonymization Model for Low-Quality Silhouettes[C]// IEEE. 2019 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC). New York: IEEE, 2019: 1686-1693.
[23]	NGOC-DUNG T T, JUNICHI Y, ISAO E. Color Transfer to Anonymized Gait Images While Maintaining Anonymization[C]// IEEE. 2020 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC). New York: IEEE, 2020: 1406-1413.
[24]	NGOC-DUNG T T, HUY H N, HOANG-QUOC N, et al. Spatio-Temporal Generative Adversarial Network for Gait Anonymization[J]. Journal of Information Security and Applications, 2019(46): 307-319.
[25]	YUKI H, KAZUAKI N, NAOKO N, et al. Anonymization of Gait Silhouette Video by Perturbing Its Phase and Shape Components[C]// IEEE. 2019 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC). New York: IEEE, 2019: 1679-1685.
[26]	JIANG Xinghao, ZHAO Zeyu, XU Ke. Adversarial Attack Technology for Vision-Based Aircraft Intelligent Object Detection[J]. Air & Space Defense, 2021, 4(1): 8-13.
	蒋兴浩, 赵泽宇, 许可. 基于视觉的飞行器智能目标检测对抗攻击技术[J]. 空天防御, 2021, 4(1): 8-13.
[27]	CHRISTIAN S, WOJCIECH Z, ILYA S, et al. Intriguing Properties of Neural Networks[EB/OL]. (2013-12-21) [2023-07-01]. https://arxiv.org/abs/1312.6199.
[28]	IAN J G, JONATHON S, CHRISTIAN S. Explaining and Harnessing Adversarial Examples[EB/OL]. (2014-12-20) [2023-07-01]. https://arxiv.org/abs/1412.6572.
[29]	ALEKSANDER M, ALEKSANDAR M, LUDWIG S, et al. Towards Deep Learning Models Resistant to Adversarial Attacks[EB/OL]. (2017-06-19) [2023-07-01]. https://arxiv.org/abs/1706.06083.
[30]	FAN Chao, LIANG Junhao, SHEN Chuanfu, et al. OpenGait: Revisiting Gait Recognition Toward Better Practicality[EB/OL]. (2022-11-12) [2023-07-01]. https://arxiv.org/abs/2211.06597.
[31]	FAN Chao, PENG Yunjie, CAO Chunshui, et al. GaitPart: Temporal Part-Based Model for Gait Recognition[C]// IEEE. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2020: 14213-14221.
[32]	LIN Beibei, ZHANG Shunli, YU Xin, et al. Gait Recognition via Effective Global-Local Feature Representation and Local Temporal Aggregation[C]// IEEE. 2021 IEEE/CVF International Conference on Computer Vision (ICCV). New York: IEEE, 2021: 14648-14656.

目标模型	方法	NM	BG	CL	加权平均
GaitBase^[30]	加权模糊*^[18]（2011）	91.20%	86.91%	70.14%	86.13%
	运动模糊*^[19]（2022）	83.95%	76.15%	55.49%	76.70%
	轮廓拉伸*^[20]（2022）	88.46%	83.57%	66.24%	83.04%
	本文方法	78.60%	70.97%	56.73%	72.70%
GaitSet^[13]	加权模糊*^[18]（2011）	85.95%	78.78%	59.75%	79.28%
	运动模糊*^[19]（2022）	77.89%	69.55%	49.39%	70.52%
	轮廓拉伸*^[20]（2022）	82.55%	77.73%	59.24%	76.92%
	本文方法	75.45%	68.31%	53.70%	69.67%
GaitPart^[31]	加权模糊*^[18]（2011）	85.29%	77.46%	63.70%	79.40%
	运动模糊*^[19]（2022）	80.57%	70.26%	54.78%	73.35%
	轮廓拉伸*^[20]（2022）	85.25%	79.45%	68.21%	80.68%
	本文方法	75.04%	66.85%	56.30%	69.65%
GaitGL^[32]	加权模糊*^[18]（2011）	90.85%	87.18%	73.76%	86.70%
	运动模糊*^[19]（2022）	89.00%	84.25%	67.39%	83.72%
	轮廓拉伸*^[20]（2022）	88.18%	85.39%	72.70%	84.52%
	本文方法	77.55%	71.57%	61.21%	73.09%

目标模型	方法	NM
GaitBase^[30]	加权模糊*^[18]（2011）	91.96%
	运动模糊*^[19]（2022）	88.06%
	轮廓拉伸*^[20]（2022）	90.98%
	本文方法	78.48%
GaitSet^[13]	加权模糊*^[18]（2011）	88.85%
	运动模糊*^[19]（2022）	87.63%
	轮廓拉伸*^[20]（2022）	87.59%
	本文方法	78.40%
GaitPart^[31]	加权模糊*^[18]（2011）	88.09%
	运动模糊*^[19]（2022）	87.33%
	轮廓拉伸*^[20]（2022）	89.93%
	本文方法	80.17%
GaitGL^[32]	加权模糊*^[18]（2011）	91.51%
	运动模糊*^[19]（2022）	90.32%
	轮廓拉伸*^[20]（2022）	92.04%
	本文方法	81.40%

$\epsilon $	NM	BG	CL	加权平均	L₂均值	SSIM
0	93.50%	88.90%	72.55%	88.39%	0	1.000
10	78.52%	71.07%	56.02%	72.53%	5521	0.986
20	77.41%	70.18%	54.40%	71.36%	6152	0.983
30	75.50%	68.14%	50.95%	69.12%	7010	0.979
40	72.58%	64.35%	46.56%	65.73%	7931	0.976
50	67.73%	57.83%	40.44%	60.29%	9036	0.972