缩略图盲解密模型：利用改进生成对抗网络的免密钥图像重建

朱礼亚; 孙雅娜; 任帅; 张刘坤; 蒋东华

doi:10.12263/DZXB.20250556

您当前的位置：

首页 >

文章列表页 >

缩略图盲解密模型：利用改进生成对抗网络的免密钥图像重建

学术论文 | 更新时间：2026-06-16

- 缩略图盲解密模型：利用改进生成对抗网络的免密钥图像重建
- A Blind Decryption Model for Thumbnail-Preserving Encryption: Image Reconstruction without a Key via an Improved Generative Adversarial Network
- 电子学报 2026年54卷第3期页码：1094-1104
- 作者机构：
  
  1.长安大学电子与控制工程学院，陕西西安 710064
  2.长安大学信息工程学院，陕西西安 710064
  3.中山大学计算机学院，广东广州 510006
- 作者简介：
  
  朱礼亚男，1980年7月出生于安徽省淮北市。现为长安大学电子与控制工程学院高级工程师、硕士生导师。主要研究方向为多媒体信息安全。E-mail: lyzhu@chd.edu.cn
  孙雅娜女，2002年2月出生于陕西省商洛市。现为长安大学电子与控制工程学院硕士研究生。主要研究方向为多媒体信息安全。E-mail: ynsun@chd.edu.cn
  任帅男，1982年9月出生于山西省太原市。现为长安大学信息工程学院教授、博士生导师。主要研究方向为信息隐藏与数字水印技术、数字取证与篡改技术、信息隐藏分析技术。E-mail: shuairen@chd.edu.cn
- 基金信息：
  
  国家自然科学基金(62372062)
- DOI：10.12263/DZXB.20250556
  中图分类号： TP391;
- 收稿：2025-06-26，
  
  录用：2026-03-06，
  
  纸质出版：2026-03-25
- 稿件说明：
移动端阅览
朱礼亚, 孙雅娜, 任帅, 等. 缩略图盲解密模型：利用改进生成对抗网络的免密钥图像重建[J]. 电子学报, 2026, 54(03): 1094-1104.

ZHU Liya, SUN Yana, REN Shuai, et al. A Blind Decryption Model for Thumbnail-Preserving Encryption: Image Reconstruction without a Key via an Improved Generative Adversarial Network[J]. Acta Electronica Sinica, 2026, 54(03): 1094-1104.
朱礼亚, 孙雅娜, 任帅, 等. 缩略图盲解密模型：利用改进生成对抗网络的免密钥图像重建[J]. 电子学报, 2026, 54(03): 1094-1104. DOI：10.12263/DZXB.20250556

ZHU Liya, SUN Yana, REN Shuai, et al. A Blind Decryption Model for Thumbnail-Preserving Encryption: Image Reconstruction without a Key via an Improved Generative Adversarial Network[J]. Acta Electronica Sinica, 2026, 54(03): 1094-1104. DOI：10.12263/DZXB.20250556

摘要

缩略图保持加密（Thumbnail-Preserving Encryption， TPE）是平衡云端图像隐私性和可用性的重要方法。为确保安全性，通常利用动态更新机制生成与原始图像信息相关联的密钥。针对图像信息非正常缺失或篡改，接收方无法获得正确的密钥进行解密的问题，设计了一种基于生成对抗网络的轻量化缩略图盲解密模型（Blind Decryption model based on Generative Adversarial Networks， BD-GAN）。通过对生成器、判别器和损失函数的改进和优化，提升解密图像的质量。生成器基于改进的U-Net网络，采用编码器-转码器-解码器的级联结构。在编码器和解码器中分别嵌入多尺度注意力融合模块（Multi-Scale Attention Fusion modules， MSAF），实现各尺度头部信息的跨层融合，在采样过程中最大限度保留图像细节，解决了多次采样和深层网络长距离依赖所导致的信息丢失问题。将U-Net网络的瓶颈层替换为由多个基础残差块级联堆叠构成的转码器，促进提取特征的高效传递与学习。为了解决局部判别器只输出最后一层局部特征，忽略中间层特征所含的多尺度语义与纹理信息的问题，改进并设计了具有多层特征反馈特性的局部判别器，返回多个中间层特征用于对抗训练。通过对不同尺度特征的逐层输出，提升判别器对多尺度纹理细节的感知能力。在损失函数优化方面，采用视觉感知损失、对抗损失和身份一致性损失的联合优化策略，将各损失项的线性组合作为优化目标，通过最小化目标函数进行训练，提升重建图像的视觉质量。实验结果表明，该模型能够对多种缩略图保持加密算法进行免密钥重建，解决了密钥有损情况下解密失败的问题。同时在不牺牲重建性能的前提下，具有更少的训练计算开销，有效地降低了部署成本。相比于采用U-Net网络和局部判别器的生成对抗网络模型，重建图像的峰值信噪比提升了0.632 8 dB，FID（Fréchet Inception Distance）性能提升了14.361 0。与去噪扩散概率解密模型相比，在保证重建效果的同时，参数量和计算量分别减少5.120 × 10

和2.807 × 10

以上。研究为通过身份验证和合法授权的用户提供了应急条件下的缩略图解密方案，提高了安全冗余度。

Abstract

Thumbnail-preserving encryption (TPE) is an important method to balance the privacy and availability of images in cloud environments. In pursuit of superior security

secret keys are generally generated using the image-related information under the assistance of a dynamic update mechanism. However

receivers may fail to obtain the correct keys due to the information loss or tampering. To address this issue

a lightweight blind decryption model based on generative adversarial networks (BD-GAN) is proposed for TPE images. This work devotes to better reconstruction performance of the image through the improvement of the generator and discriminator

and optimization of the loss function. In our scheme

the generator is designed based on an improved U-Net architecture

and it employs a cascaded structure composed of encoder

tra

nscoder

and decoder modules. Multi-scale attention fusion modules (MSAF) are embedded in the encoder and decoder

respectively

to achieve cross-layer fusion of multi-scale hierarchical information. By this means

image details are reserved to the maximum extent

and the problems of information loss caused by repeated sampling and long-range dependencies in deep networks are addressed. The bottleneck layer of the U-Net network is replaced with a transcoder composed of cascaded multiple residual blocks

so as to promote the efficient transmission and learning of extracted features. Moreover

to solve the problem that the local discriminator only utilizes the local features of the last layer and ignores the multi-scale semantic and texture information contained in the intermediate layer features

a local discriminator with multi-layer feature feedback is designed

which returns multiple intermediate layer features for adversarial training. Through the layer-by-layer output of features

the capability of perceiving the multi-scale textures is enhanced. In addition

a joint optimization strategy integrating visual perception loss

adversarial loss

and identity consistency loss is adopted to optimize the loss function. Specifically

the linear combination of these loss components serves as the optimization objective

and the model is trained by minimizing the overall loss to improve the visual quality of the reconstructed images. Experimental results show that the proposed model can perform for various TPE algorithms without any secret keys

and the failure of decryption can be avoided even if the secret keys are damaged. More importantly

on the premise of not degrading reconstruction performance

our model has lower training computational overhead and deployment cost. Compared with the traditional GAN employing a U-Net network and local discriminator

the peak signal to noise ratio (PSNR) values of the reconstructed images are improved by 0.632 8 dB

and the fréchet inception distance (FID) values are improved by

14.361 0. Notably

in comparison with the diffusion models

our model offers a satisfactory reconstructed image while reducing parameters and computational cost by more than 5.120 × 10

and 2.807 × 10

respectively. This study provides an effective emergency decryption scheme for users who have passed identity verification and legal authorization

which improves security redundancy.

关键词

Keywords

references

Kagan D M , Alpert G F , Fire M . Zooming into video conferencing privacy [J ] . IEEE Transactions on Computational Social Systems , 2024 , 11 ( 1 ): 933 - 944 . DOI: 10.1109/tcss.2022.3231987 http://dx.doi.org/10.1109/tcss.2022.3231987

Fainmesser I P , Galeotti A , Momot R . Digital privacy [J ] . Management Science , 2023 , 69 ( 6 ): 3157 - 3173 . DOI: 10.1287/mnsc.2022.4513 http://dx.doi.org/10.1287/mnsc.2022.4513

Wen W Y , Yuan Z Y , Qi S R , et al . PPM-SEM: A privacy-preserving mechanism for sharing electronic patient records and medical images in telemedicine [J ] . IEEE Transactions on Multimedia , 2024 , 26 : 5795 - 5806 . DOI: 10.1109/tmm.2023.3339588 http://dx.doi.org/10.1109/tmm.2023.3339588

宋昭阳 , 王一诺 , 王浩文 , 等 . 基于Hopfield网络“伪吸引子”与交替量子随机行走的抗攻击彩色图像加密方案 [J ] . 电子学报 , 2023 , 51 ( 8 ): 2030 - 2042 .

Song Zhaoyang , Wang Yinuo , Wang Haowen , et al . Anti-attack color image encryption scheme based on Hopfield network “pseudo attractor” and alternating quantum random walk [J ] . Acta Electronica Sinica , 2023 , 51 ( 8 ): 2030 - 2042 . (in Chinese)

Song J C , Chen B , Zhang J . Dynamic path-controllable deep unfolding network for compressive sensing [J ] . IEEE Transactions on Image Processing , 2023 , 32 : 2202 - 2214 . DOI: 10.1109/tip.2023.3263100 http://dx.doi.org/10.1109/tip.2023.3263100

陈云 , 罗成 , 许璐 . 基于两种置乱方式的混合混沌图像加密算法 [J ] . 通信技术 , 2024 , 57 ( 9 ): 925 - 933 .

Chen Yun , Luo Cheng , Xu Lu . A hybrid chaotic image encryption algorithm based on two scrambling methods [J ] . Communications Technology , 2024 , 57 ( 9 ): 925 - 933 . (in Chinese)

Jiang D H , Njitacke Z T , Long G Q , et al . Novel Tabu learning neuron model with variable activation gradient and its application to secure healthcare [J ] . Chaos , Solitons & Fractals, 2024 , 189 : 115632 . DOI: 10.1016/j.chaos.2024.115632 http://dx.doi.org/10.1016/j.chaos.2024.115632

Wright C V , Feng W C , Liu F . Thumbnail-preserving encryption for JPEG [C ] // Proceedings of the 3rd ACM Workshop on Information Hiding and Multimedia Security . New York : ACM , 2015 : 141 - 146 . DOI: 10.1145/2756601.2756618 http://dx.doi.org/10.1145/2756601.2756618

Tajik K , Gunasekaran A , Dutta R , et al . Balancing image privacy and usability with thumbnail-preserving encryption [C ] // Proceedings 2019 Network and Distributed System Security Symposium . Internet Society , 2019 . DOI: 10.14722/ndss.2019.23432 http://dx.doi.org/10.14722/ndss.2019.23432

Zhao R Y , Zhang Y S , Xiao X L , et al . TPE2: Three-pixel exact thumbnail-preserving image encryption [J ] . Signal Processing , 2021 , 183 : 108019 . DOI: 10.1016/j.sigpro.2021.108019 http://dx.doi.org/10.1016/j.sigpro.2021.108019

Zhang Y S , Zhou W T , Zhao R Y , et al . F-TPE: Flexible thumbnail-preserving encryption based on multi-pixel sum-preserving encryption [J ] . IEEE Transactions on Multimedia , 2023 , 25 : 5877 - 5891 . DOI: 10.1109/tmm.2022.3200310 http://dx.doi.org/10.1109/tmm.2022.3200310

Marohn B , Wright C V , Feng W C , et al . Approximate thumbnail preserving encryption [C ] // Proceedings of the 2017 on Multimedia Privacy and Security . New York : ACM , 2017 : 33 - 43 . DOI: 10.1145/3137616.3137621 http://dx.doi.org/10.1145/3137616.3137621

Zhang Y S , Zhao R Y , Xiao X L , et al . HF-TPE: High-fidelity thumbnail- preserving encryption [J ] . IEEE Transactions on Circuits and Systems for Video Technology , 2022 , 32 ( 3 ): 947 - 961 . DOI: 10.1109/tcsvt.2021.3070348 http://dx.doi.org/10.1109/tcsvt.2021.3070348

Chai X L , Ma Y K , Wang Y J , et al . TPE-ADE: Thumbnail-preserving encryption based on adaptive deviation embedding for JPEG images [J ] . IEEE Transactions on Multimedia , 2024 , 26 : 6102 - 6116 . DOI: 10.1109/tmm.2023.3345158 http://dx.doi.org/10.1109/tmm.2023.3345158

Wang X , Qu L F , Wu H T , et al . Reversible image thumbnail preservation with high-visual naturalness [J ] . IEEE Internet of Things Journal , 2024 , 11 ( 10 ): 18739 - 18752 . DOI: 10.1109/jiot.2024.3366328 http://dx.doi.org/10.1109/jiot.2024.3366328

赵若宇 , 叶茜 , 周文韬 , 等 . 云存储图像缩略图保持的加密研究进展 [J ] . 中国图象图形学报 , 2023 , 28 ( 3 ): 645 - 665 . DOI: 10.11834/jig.220533 http://dx.doi.org/10.11834/jig.220533

ZHAO Ruoyu , YE Xi , ZHOU Wentao , et al . Cloud-stored image thumbnail-preserving encryption [J ] . Journal of Image and Graphics , 2023 , 28 ( 3 ): 645 - 665 . (in Chinese) . DOI: 10.11834/jig.220533 http://dx.doi.org/10.11834/jig.220533

Li Z Y , Xie D , Liu S Q , et al . Known-plaintext attacks to thumbnail-preservation encryption using Pix2pix generative adversarial network [C ] // ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing . Piscataway : IEEE , 2025 : 10890779 . DOI: 10.1109/icassp49660.2025.10890779 http://dx.doi.org/10.1109/icassp49660.2025.10890779

Ledig C , Theis L , Huszár F , et al . Photo-realistic single image super-resolution using a generative adversarial network [C ] // 2017 IEEE Conference on Computer Vision and Pattern Recognition . Piscataway : IEEE , 2017 : 105 - 114 . DOI: 10.1109/cvpr.2017.19 http://dx.doi.org/10.1109/cvpr.2017.19

Zhang P Y , Zhang K H , Luo W H , et al . Blind face restoration: Benchmark datasets and a baseline model [J ] . Neurocomputing , 2024 , 574 : 127271 . DOI: 10.1016/j.neucom.2024.127271 http://dx.doi.org/10.1016/j.neucom.2024.127271

Saharia C , Ho J , Chan W , et al . Image super-resolution via iterative refinement [J ] . IEEE Transactions on Pattern Analysis and Machine Intelligence , 2023 , 45 ( 4 ): 4713 - 4726 .

Jiang D H , Ni J Q , Alasbali N , et al . DDCM: Cracking anonymized facial images using denoising diffusion cryptanalytic model [J ] . IEEE Transactions on Consumer Electronics , 2025 , 71 ( 2 ): 4464 - 4474 . DOI: 10.1109/tce.2025.3534859 http://dx.doi.org/10.1109/tce.2025.3534859

Bellare M , Ristenpart T , Rogaway P , et al . Format-preserving encryption [M ] // Selected areas in cryptography . Berlin, Heidelberg : Springer , 2009 : 295 - 312 . DOI: 10.1007/978-3-642-05445-7_19 http://dx.doi.org/10.1007/978-3-642-05445-7_19

Goodfellow I J , Pouget-Abadie J , Mirza M , et al . Generative adversarial nets [C ] // Proceedings of the 28th International Conference on Neural Information Processing Systems - Volume 2 . New York : ACM , 2014 : 2672 - 2680 .

Vaswani A , Shazeer N , Parmar N , et al . Attention is all you need [J ] . Advances in Neural Information Processing Systems , 2017 , 30 : 5998 - 6008 .

Kim J , Nang J , Choe J . LMLT: Low-to-high multi-level vision transformer for lightweight image super- resolution [C ] // 2025 IEEE/CVF International Conference on Computer Vision Workshops . Piscataway : IEEE , 2025 : 5568 - 5578 . DOI: 10.1109/iccvw69036.2025.00583 http://dx.doi.org/10.1109/iccvw69036.2025.00583

Ronneberger O , Fischer P , Brox T . U-Net: Convolutional networks for biomedical image segmentation [M ] // Medical image computing and computer-assisted intervention - MICCAI 2015 . Cham : Springer International Publishing , 2015 : 234 - 241 . DOI: 10.1007/978-3-319-24574-4_28 http://dx.doi.org/10.1007/978-3-319-24574-4_28

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 . Piscataway : IEEE , 2017 : 5967 - 5976 . DOI: 10.1109/cvpr.2017.632 http://dx.doi.org/10.1109/cvpr.2017.632

Johnson J , Alahi A , Li F F . Perceptual losses for real-time style transfer and super-resolution [M ] // Computer vision - ECCV 2016 . Cham : Springer International Publishing , 2016 : 694 - 711 . DOI: 10.1007/978-3-319-46475-6_43 http://dx.doi.org/10.1007/978-3-319-46475-6_43

Schroff F , Kalenichenko D , Philbin J . FaceNet: A unified embedding for face recognition and clustering [C ] // 2015 IEEE Conference on Computer Vision and Pattern Recognition . Piscataway : IEEE , 2015 : 815 - 823 . DOI: 10.1109/cvpr.2015.7298682 http://dx.doi.org/10.1109/cvpr.2015.7298682

Parkhi O M , Vedaldi A , Zisserman A . Deep face recognition [C ] // Proceedings ofthe British Machine Vision Conference 2015 . British Machine Vision Association , 2015 : C.29.41 . DOI: 10.5244/c.29.41 http://dx.doi.org/10.5244/c.29.41

Heusel M , Ramsauer H , Unterthiner T , et al . GANs trained by a two time-scale update rule converge to a local Nash equilibrium [C ] // Proceedings of the 31st International Conference on Neural Information Processing Systems . New York : ACM , 2017 : 6629 - 6640 . DOI: 10.48550/arXiv.1706.08500 http://dx.doi.org/10.48550/arXiv.1706.08500

Liu X R , Meng X F , Wang Y R , et al . Known-plaintext cryptanalysis for a computational-ghost-imaging cryptosystem via the Pix2Pix generative adversarial network [J ] . Optics Express , 2021 , 29 ( 26 ): 43860 . DOI: 10.1364/oe.445689 http://dx.doi.org/10.1364/oe.445689

浏览量

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

特征掩码与对比学习融合多维度去过度相关的序列推荐

基于GAN-Data数据增强与FD-DETR的织物疵点检测方法

基于注意力机制优化的生成对抗网络及其在海杂波模拟中的应用

天波超视距雷达地海杂波图像增强与检测器设计

DoS攻击下多无人机群体智能系统分布式一致性跟踪控制