基于压缩感知和生成网络的多层级可逆人脸隐私保护  

Multi⁃Level Reversible Facial Privacy Protection Based on Compressive Sensing and Generative Adversarial Network

作　　者：高献伟[1] 程逸煊 索珠峰 GAO Xianwei;CHENG Yixuan;SUO Zhufeng(Beijing Electronics Science and Technology Institute,Beijing 100070,P.R.China;Beijing University of Post and Telecommunications,Beijing 100876,P.R.China)

机构地区：[1]北京电子科技学院,北京100070 [2]北京邮电大学,北京100876

出　　处：《北京电子科技学院学报》2024年第4期18-29,共12页Journal of Beijing Electronic Science And Technology Institute

基　　金：中央高校基本科研业务费资金资助(项目编号:328xxxxxx)。

摘　　要：云平台是一种先进的现代化的信息传送、存储及共享服务,它的主要挑战之一是确保用户数据的安全性、保密性、完整性和可用性。为避免数据泄露潜在损害,本文提出了一种基于压缩感知和生成网络的多层级可逆人脸隐私保护方案,该方案通过双重风格生成对抗网络技术将原始图像人脸生成动漫人脸,进而将动漫人脸嵌入到原始图像生成秘密图像,接着对秘密图像压缩采样,再嵌入水印,实现对整个图像加密,最后生成密文图像。完全授权用户可以提取水印,恢复原始图像,受限授权用户只能恢复秘密图像。所提出的人脸隐私保护方案提供了一种低成本的加密系统,仿真结果表明,该系统的计算复杂度和时间复杂度低,重构质量高,运行时间短。The cloud platform is an advanced and modernized information delivery,storage and sharing service,and one of its main challenges is to ensure the security,confidentiality,integrity and availabil⁃ity of user data.In order to avoid the potential damage of data leakage,in this paper,we propose a multi⁃level reversible face privacy preservation system based on compressive sensing and generative ad⁃versarial network.The system generates the face of the original image into the face of the anime through the dual style generative adversarial network,embeds the anime face into the original image to generate the secret image.Compressive sensing provides the encryption of the secret image,and the watermark is embedded in the encrypted compressive sensing measurement values,finally generates a cipher image.Fully authorized users can extract the watermark and recover the original image,while semi⁃authorized users can only recover the secret image.Extensive experimental results indicate the proposed facial pri⁃vacy preservation method provides a low⁃cost encryption system with low computational complexity and time complexity.While achieving slightly higher reconstruction quality,the runtime of competing system is nearly twice that of the system presented in this paper.

关 键 词：压缩感知 生成网络 多层级 加密 可逆隐私保护 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]