A high-capacity image steganography algorithm based on end-to-end deep learning networks
-
摘要:
图像具有纹理复杂、冗余空间大、应用普遍广泛等特点，以图像为载体的隐写算法是隐写术的主流方向。近年来，基于深度学习网络的图像隐写算法获得了越来越好的隐写性能，成为图像隐写领域的研究热点。基于此，基于生成对抗网络(GAN)提出了一种大容量的图像隐写算法。算法设计了以预处理模块、卷积神经网络(CNN)模块和U-Net为主体的信息嵌入提取网络，通过损失函数的约束，联合训练嵌入、提取和判别网络，实现了良好的隐写视觉效果。实验结果表明：所提算法的嵌入容量达到24 bit/像素。在此大容量嵌入的前提下，所提算法生成的载密图像和提取的秘密图像无论在主观视觉质量还是客观视觉指标峰值信噪比(PSNR)上都高于其他同类算法，说明了设计的端到端隐写网络的整体优越性。
Abstract:Due to the complex texture, large redundant space, and widespread application of images, image based steganography algorithms are still the mainstream direction of steganography. Deep neural network-based picture steganography algorithms have become a research hotspot in the field of image steganography in recent years due to their increasingly good steganographic performance. This article proposes a high-capacity image steganography algorithm based on a generative adversarial network (GAN). The algorithm designed an information embedding and extraction network with a preprocessing module, a convolutional neural networks (CNN) module, and a U-Net as the main components. Through the constraint of the loss function, the embedding, extraction, and discrimination networks were jointly trained to achieve good steganographic visual effects. The experimental results show that our algorithm achieves an embedding capacity of 24 bpp. The overall superiority of the end-to-end steganography network designed in this paper is demonstrated by the fact that, under the assumption of high-capacity embedding, the encrypted images produced by the algorithm in this paper and the extracted secret images are higher than other comparable algorithms in both subjective visual quality and objective visual indicator peak signal-to-noise ratio (PSNR).
-
Key words:
- image steganography /
- deep learning /
- GNN /
- U-Net /
- information hiding
-
表 1 U-Net详细参数配置
Table 1. U-Net detailed parameter configuration
输入维度大小/
像素×像素×像素处理方式 卷积核
大小步长 填充 输出维度大小/
像素×像素×像素256×256×32 输入/输出 256×256×32 卷积 3×3 1 1 复制维度拼接 最大池化 2 上采样 2×2 2 0 
下载: 导出CSV
表 2 嵌入式网络详细参数配置
Table 2. Detailed parameter configuration of embedding network
输入维度大小 处理方式 卷积核大小 通道数 输出维度大小 256×256×3 预处理 3×3 1 256×256×1 256×256×3 预处理 5×5 1 256×256×1 256×256×3 预处理 7×7 1 256×256×1 256×256×1 预处理 3×3 3 256×256×3 256×256×1 预处理 3×3 3 256×256×3 256×256×1 预处理 3×3 3 256×256×3 256×256×3 卷积模块 3×3 8 256×256×8 256×256×8 维度拼接 3×3 16 256×256×16 256×256×8 维度拼接 3×3 16 256×256×16 256×256×16 卷积模块 4×4 32 255×255×32 255×255×32 卷积模块 2×2 32 256×256×32 256×256×32 U-Net 32 256×256×32 255×255×64 卷积模块 3×3 3 256×256×3 256×256×3 残差模块 256×256×3
下载: 导出CSV
表 3 提取网络的详细参数配置
Table 3. Detailed parameter configuration of Extraction Network
输入维度大小/
像素×像素×像素处理
方式卷积核
大小通道数 输出维度大小/
像素×像素×像素256×256×3 卷积模块 3×3 32 256×256×32 256×256×32 U-Net 32 256×256×32 256×256×32 卷积模块 4×4 8 255×255×8 255×255×8 卷积模块 2×2 3 256×256×3 256×256×3 卷积模块 3×3 3 256×256×3
下载: 导出CSV
表 4 隐写算法的单位嵌入容量对比
Table 4. Comparison of unit embedding capacity of steganography algorithms
下载: 导出CSV
-
[1] Wang C, Ni J. An efficient JPEG steganographic scheme based on the block entropy of DCT coefficients[C]//Proceedings of the 2012 IEEE International Conference on Acoustics, Speech and Signal Processing. Piscataway: IEEE Press, 2012: 1785-1788. [2] Guo L J, Ni J Q, Shi Y Q. Uniform embedding for efficient JPEG steganography[J]. IEEE transactions on Information Forensics and Security, 2014, 9(5): 814-825. [3] Guo L J, Ni J Q, Su W K, et al. Using statistical image model for JPEG steganography: uniform embedding revisited[J]. IEEE Transactions on Information Forensics and Security, 2015, 10(12): 2669-2680. [4] Salunkhe S, Bhosale S, Narkhede S V. An efficient video steganography for pixel location optimization using Fr-WEWO algorithm based deep CNN model[J]. International Journal of Image, Graphics and Signal Processing, 2023, 15(3): 14-30. [5] Khandelwal J, Sharma V K. Extensive dual tree complex wavelet transform-based image steganography using SVD and CNN subspace[J]. Journal of Discrete Mathematical Sciences and Cryptography, 2023, 26(3): 617-627. [6] Wang Z H, Byrnes O, Wang H, et al. Data hiding with deep learning: a survey unifying digital watermarking and steganography[J]. IEEE Transactions on Computational Social Systems, 2023, 10(6): 2985-2999. [7] Rehman H A, Bajwa U I, Raza R H, et al. Leveraging coverless image steganography to hide secret information by generating anime characters using GAN[J]. Expert Systems with Applications, 2024, 248: 123420. [8] Ahmadi M, Norouzi A, Karimi N, et al. ReDMark: framework for residual diffusion watermarking based on deep networks[J]. Expert Systems with Applications, 2020, 146: 113157. [9] Liu J, Zhou T P, Zhang Z, et al. Digital cardan grille: a modern approach for information hiding[C]//Proceedings of the 2018 2nd International Conference on Computer Science and Artificial Intelligence. New York: ACM, 2018: 441-446. [10] Cui Q, Zhou Z L, Fu Z J, et al. Image steganography based on foreground object generation by generative adversarial networks in mobile edge computing with Internet of Things[J]. IEEE Access, 2019, 7: 90815-90824. [11] Volkhonskiy D, Nazarov I, Burnaev E. Steganographic generative adversarial networks[C]//Proceedings of the Twelfth International Conference on Machine Vision. Amsterdam: SPIE, 2020: 97. [12] Hayes J, Danezis G. Generating steganographic images via adversarial training[C]//Proceedings of the Advances in Neural Information Processing Systems. Long Beach: Curran Associates Ins., 2017: 30. [13] Zhu J R, Kaplan R, Johnson J, et al. HiDDeN: hiding data with deep networks[M]//Proceedings of the Computer Vision-ECCV 2018. Berlin: Springer, 2018: 682-697. [14] Wang Z H, Gao N, Wang X, et al. SSteGAN: self-learning steganography based on generative adversarial networks[C]//Proceedings of the Neural Information Processing. Berlin: Springer, 2018: 253-264. [15] Wu P, Yang Y, Li X Q. StegNet: mega image steganography capacity with deep convolutional network[J]. Future Internet, 2018, 10(6): 54. [16] Subramanian N, Cheheb I, Elharrouss O, et al. End-to-end image steganography using deep convolutional autoencoders[J]. IEEE Access, 2021, 9: 135585-135593. [17] Ronneberger O, Fischer P, Brox T. U-Net: convolutional networks for biomedical image segmentation[C]//Proceedings of the Medical Image Computing and Computer-Assisted Intervention–MICCAI 2015. Berlin: Springer, 2015, 9351: 234-241. [18] Kingma D P, Ba J. Adam: a method for stochastic optimization[EB/OL]. (2017-01-30)[2024-03-06]. https://doi.org/10.48550/arXiv.1412.6980. [19] Lin T Y, Maire M, Belongie S, et al. Microsoft COCO: Common objects in context[C]//Proceedings of the Computer Vision-ECCV 2014. Berlin: Springer, 2014: 740-755. [20] Zhang K A, Cuesta-infante A, Xu L, et al. SteganoGAN: high capacity image steganography with GANs[EB/OL]. (2019-01-30)[2024-03-10]. https://doi.org/10.48550/arXiv.1901.03892. [21] Ur Rehman A, Rahim R, Nadeem S, et al. End-to-end trained CNN encoder-decoder networks for image steganography[C]//Proceedings of the Computer Vision-ECCV 2018 Workshops. Berlin: Springer, 2019: 723-729. -
下载:
点击查看大图
计量
- 文章访问数: 389
- HTML全文浏览量: 122
- PDF下载量: 30
- 被引次数: 0
