社交媒体和云平台为图像的传播和存储带来了便利，但同时也引起了人们对于图像隐私的担忧。因此，需要采取一定的措施去保护图像的隐私，以防止隐私被窃取和非法使用。基于上述目标，本文提出了基于循环对抗网络（cycle-consistent generative adversarialnetworks,CycleGAN）的图像隐私保护。为了在图像隐私保护中兼顾可用性，该方法先用图像分割和CycleGAN组合，选择出不同的分割系数来辅助生成不同程度的隐私保护图像。然后利用可逆信息隐藏对生成的隐私保护图像进行信息的嵌入，从而阻止非法使用者在图像重构中提取隐私信息，进而保证了整个过程图像隐私保护和可用性的平衡。本文用PIPA数据集对该方法进行训练和测试，采用峰值信噪比和结构相似性指数作为客观指标对隐私保护的图像进行评估。实验结果表明，本方案在图像隐私保护和可用性两方面都优于其他对比方案。
Social media and cloud computing have facilitated the distribution and storage of images. Meanwhile, concerns about image privacy have been raised. It is crucial to protect image privacy from privacy violation and illegal use. Motivated by this, an image privacy protection method based on cycle-consistent generative adversarial networks (CycleGAN) is proposed in this paper. Considering the usability in image privacy protection, the method first combines image segmentation with CycleGAN to select different segmentation coefficients to generate images with different degrees of privacy protection. Then reversible information hiding is used to embed information in the generated privacy preserving image, so as to prevent illegal users from reversing the image. Thus, a balance is achieved between image privacy protection and usability. The proposed method is trained and tested using PIPA dataset, using peak signal to noise ratio and structural similarity index are used as performance metrics to evaluate the privacy-preserving images. Experimental results show that the proposed method outperforms other schemes in both image privacy preservation and usability.
[1] Mcpherson R, Shokri R, Shmatikov V. Defeating image obfuscation with deep learning[DB/OL]. 2016[2022-10-29]. https://arxiv.org/abs/1609.00408.
[2] Raval N, Machanavajjhala A, Cox L P. Protecting visual secrets using adversarial nets[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2017:1329-1332.
[3] Nandakumar K, Ratha N, Pankanti S, et al. Towards deep neural network training on encrypted data[C]//2019 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2019:40-48.
[4] 张赛男, 李千目. 彩色图像隐私部分自动标定及加密[J]. 南京理工大学学报, 2022, 46(4):395-405. Zhang S N, Li Q M. Automatic calibration and encryption of privacy part of color image[J]. Journal of Nanjing University of Science and Technology, 2022, 46(4):395-405. (in Chinese)
[5] Newton E M, Sweeney L, Malin B. Preserving privacy by de-identifying face images[J]. IEEE Transactions on Knowledge and Data Engineering, 2005, 17(2):232-243.
[6] Gross R, Airoldi E, Malin B, et al. Integrating utility into face de-identification[C]//International Workshop on Privacy Enhancing Technologies. Springer, 2005:227-242.
[7] Meng L, Shenoy A. Retaining expression on de-identified faces[C]//International Conference on Speech and Computer. Springer, 2017:651-661.
[8] Tabassum A, Erbad A, Lebda W, et al. FEDGAN-IDS:privacy-preserving IDS using GAN and federated learning[J]. Computer Communications, 2022, 192:299-310.
[9] 葛瑞. 基于生成对抗网络的图像隐私保护方法研究[D]. 西安:西安电子科技大学, 2020.
[10] Ren Z Z, Lee Y J, Ryoo M S. Learning to anonymize faces for privacy preserving action detection[C]//European Conference on Computer Vision, 2018:639-655.
[11] Wu Y F, Yang F, Xu Y, et al. Privacy-protective-GAN for privacy preserving face deidentification[J]. Journal of Computer Science and Technology, 2019, 34(1):47-60.
[12] 毛典辉, 李子沁, 蔡强, 等. 基于DCGAN反馈的深度差分隐私保护方法[J]. 北京工业大学学报, 2018(6):870-877. Mao D H, Li Z Q, Cai Q, et al. Tickling deep differential privacy protection method based on DCGAN[J]. Journal of Beijing University of Technology, 2018(6):870-877. (in Chinese)
[13] 周莉莉, 姜枫. 图像分割方法综述研究[J]. 计算机应用研究, 2017, 34(7):1921-1928. Zhou L L, Jiang F. Survey on image segmentation methods[J]. Application Research of Computers, 2017, 34(7):1921-1928. (in Chinese)
[14] Shelhamer E, Long J, Darrell T. Fully convolutional networks for semantic segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(4):640-651.
[15] Borse S, Wang Y, Zhang Y Z, et al. InverseForm:a loss function for structured boundaryaware segmentation[C]//2021 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2021:5897-5907.
[16] Felzenszwalb P F, Huttenlocher D P. Efficient graph-based image segmentation[J]. International Journal of Computer Vision, 2004, 59(2):167-181.
[17] Goodfellow I, Pouget-Abadie J, Mirza M, et al. Generative adversarial nets[J]. Advances in Neural Information Processing Systems, 2014:2672-2680.
[18] Men Y F, Mao Y M, Jiang Y N, et al. Controllable person image synthesis with attributedecomposed GAN[C]//2020 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2020:5083-5092.
[19] Zhu J Y, Park T, Isola P, et al. Unpaired image-to-image translation using cycle-consistent adversarial networks[C]//2017 IEEE International Conference on Computer Vision, 2017:2242-2251.
[20] Isola P, Zhu J Y, Zhou T H, et al. Image-to-image translation with conditional adversarial networks[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017:5967-5976.
[21] Puteaux P, Ong S Y, Wong K S, et al. A survey of reversible data hiding in encrypted images-the first 12 years[J]. Journal of Visual Communication and Image Representation, 2021, 77:103085.
[22] Abadi M, Andersen D G. Learning to protect communications with adversarial neural cryptography[DB/OL]. 2016[2022-10-29]. https://arxiv.org/abs/1610.06918.
[23] Chu C, Zhmoginov A, Sandler M. Cyclegan, a master of steganography[DB/OL]. 2017[2022-10-29]. https://arxiv.org/abs/1712.02950.
[24] Wu H, Tian X J, Li M H, et al. PECAM:privacy-enhanced video streaming and analytics via securely-reversible transformation[C]//27th Annual International Conference on Mobile Computing and Networking, 2021:229-241.
[25] Kim H J, Sachnev V, Shi Y Q, et al. A novel difference expansion transform for reversible data embedding[J]. IEEE Transactions on Information Forensics and Security, 2008, 3(3):456-465.
[26] Zhang N, Paluri M, Taigman Y, et al. Beyond frontal faces:improving Person Recognition using multiple cues[C]//2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015:4804-4813.
