收稿日期: 2024-02-06
网络出版日期: 2024-05-18
基金资助
国家自然科学基金(U22B2047,62202310,U23B2022);中国博士后科学基金(2022M722192)
版权
Scaling-based image-level feature enhancement for steganalysis
Received date: 2024-02-06
Online published: 2024-05-18
Supported by
Natural Science Foundation of China (U22B2047, 62202310, U23B2022), China Postdoctoral Science Foundation (2022M722192), Shenzhen R&D Program (JCYJ20200109105008228).
Copyright
随着深度学习的快速发展,基于深度学习的图像隐写分析技术研究取得了显著进展。然而,在残差特征提取及增强方面,传统图像预处理增强技术往往导致隐写信号的减弱,使得简单的图像预处理方法难以适配于隐写分析。对此,现有的深度学习隐写分析研究倾向于在不损害图像原有信息的基础上,设计固定的滤波核或对残差卷积层优化学习,缺乏对图像层面的隐写特征增强策略的可行性探讨。针对这一现象,提出了一种新颖高效的图像级特征增强隐写分析方法,通过最近邻插值算法扩大图像尺寸,在保留原始隐写信号的基础上进一步拓展分布相同的嵌入信号,从而增强模型对隐写残差特征的提取能力,无须对现有隐写分析流程做出显著改动即可有效提高隐写痕迹的可检测性。实验结果显示,所提方法能够显著提升模型在多种隐写算法下的检测准确率,尤其对于低嵌入率,其准确率最高可提升2.81%。该方法证实了图像层面预处理在隐写残差特征增强上的有效性,为深度学习隐写分析的图像残差特征提取提供了新的研究视角。
刘绪龙 , 李伟祥 , 林凯清 , 李斌 . 基于尺寸变换的图像级特征增强隐写分析方法[J]. 网络空间安全科学学报, 2024 , 2(1) : 101 -112 . DOI: 10.20172/j.issn.2097-3136.240109
With the rapid development of deep learning, the research of image steganalysis techniques based on deep learning have made significant progress. However, in terms of residual feature extraction and enhancement, traditional image preprocessing enhancement techniques often inevitably weaken the steganographic signals, making it difficult to adapt simple image preprocessing methods to steganalysis. Therefore, existing deep learning steganalysis research tends to design fixed filter kernels or optimize the learning of the residual convolutional layers, resulting in a relative lack of exploration of steganographic feature enhancement at the input image level. In this regard, a novel and efficient method for image-level feature enhancement in steganalysis is proposed. By employing the nearest neighbor interpolation algorithm to expand the size of image, the image’ s steganographic signals is amplified while maintaining their original distribution. This further enhances the model’ s capability of steganographic residual feature extraction, and effectively improves the detectability of steganographic traces without making significant changes to the existing steganalysis process. The experimental results show that the proposed method can significantly improve the model’ s detection accuracy under various steganography algorithms, especially for low embedding rate environment where the accuracy can be improved by 2.81%. It confirms the effectiveness of image-level preprocessing on steganographic residual feature enhancement, and provides a new research perspective on image residual feature extraction for deep learning steganalysis models.
表 1 在BOSSBase图像库、S-UNIWARD隐写算法和0.2/0.4 bpp嵌入率下,不同缩放尺寸对 LWENet 检测准确率的影响Table 1 Effect of different scaling sizes on LWENet detection accuracy under BOSSBase dataset, S-UNIWARD steganography algorithm and 0.2/0.4 bpp embedding rates |
| 嵌入率/bpp | 准确率 | |||||||
| 256×256(原始) | 64×64 | 128×128 | 384×384 | 512×256 | 256×512 | 512×512 | 768×768 | |
| 0.2 | 77.72% | 52.28% | 55.97% | 78.23% | 80.57% | 79.97% | 80.53% | 80.87% |
| 0.4 | 89.42% | 54.93% | 63.32% | 87.80% | 88.95% | 88.75% | 90.45% | 89.80% |
表 2 在BOSSBase图像库和0.1/0.2/0.3/0.4 bpp嵌入率下,不同模型对原始图像与放大图像的检测准确率Table 2 Detection accuracy of different models for original and enlarged images under BOSSBase dataset and 0.1/0.2/0.3/0.4 bpp embedding rates |
| 隐写算法 | 图像 | 准确率 | |||||||
| CovNet | LWENet | ||||||||
| 0.1 | 0.2 | 0.3 | 0.4 | 0.1 | 0.2 | 0.3 | 0.4 | ||
| S-UNIWARD | 原始 | 66.00% | 80.07% | 85.65% | 89.65% | 70.13% | 77.72% | 84.82% | 89.42% |
| 放大 | 70.47% | 81.47% | 85.88% | 89.32% | 71.65% | 80.53% | 86.02% | 90.45% | |
| HILL | 原始 | 66.63% | 75.42% | 79.67% | 84.38% | 67.57% | 75.87% | 79.50% | 85.57% |
| 放大 | 68.47% | 76.55% | 81.27% | 84.78% | 67.93% | 77.42% | 81.20% | 86.19% | |
| MiPOD | 原始 | 64.52% | 74.98% | 78.95% | 83.27% | 65.28% | 72.52% | 78.67% | 84.72% |
| 放大 | 66.93% | 75.58% | 80.43% | 85.30% | 65.65% | 74.70% | 80.63% | 85.00% | |
表 3 在BOSSBase图像库、S-UNIWARD隐写算法和0.2/0.4 bpp嵌入率下,不同隐写分析模型的检测准确率Table 3 Detection accuracy of different steganalysis models under BOSSBase dataset, S-UNIWARD steganography algorithm and 0.2/0.4 bpp embedding rates |
| 隐写分析模型 | 准确率 | |
| 0.2 | 0.4 | |
| SRNet | 79.65% | 89.25% |
| CovNet + 所提方法 | 81.47% | 89.32% |
| LWENet + 所提方法 | 80.53% | 90.45% |
表 4 在BOSSBase图像库和0.1/0.2/0.3/0.4 bpp嵌入率下,CovNet对不同尺寸图像的检测准确率Table 4 Detection accuracy of CovNet for images of different sizes under BOSSBase dataset and 0.1/0.2/0.3/0.4 bpp embedding rates |
| 算法 | 图像 | 准确率 | |||
| 0.1 | 0.2 | 0.3 | 0.4 | ||
| S-UNIWARD | 原始 | 59.92% | 69.63% | 76.68% | 82.70% |
| 放大至2倍 | 60.90% | 71.57% | 78.47% | 83.27% | |
| 放大至4倍 | 61.12% | 71.17% | 77.80% | 83.33% | |
| HILL | 原始 | 59.07% | 66.33% | 72.68% | 78.17% |
| 放大至2倍 | 60.98% | 68.85% | 75.28% | 78.97% | |
| 放大至4倍 | 61.70% | 69.60% | 74.32% | 78.65% | |
| MiPOD | 原始 | 58.75% | 67.07% | 72.38% | 77.22% |
| 放大至2倍 | 59.85% | 67.75% | 73.32% | 77.68% | |
| 放大至4倍 | 60.02% | 68.15% | 72.05% | 78.20% | |
表 5 在SZUBase图像库和0.1/0.2/0.3/0.4 bpp嵌入率下,CovNet对不同隐写算法和嵌入率图像的检测准确率Table 5 Detection accuracy of CovNet for images with different steganography algorithms and embedding rates under SZUBase dataset and 0.1/0.2/0.3/0.4 bpp embedding rates |
| 算法 | 图像 | 准确率 | |||
| 0.1 | 0.2 | 0.3 | 0.4 | ||
| S-UNIWARD | 原始 | 67.05% | 77.78% | 82.73% | 86.65 |
| 放大 | 68.47% | 77.62% | 83.35% | 86.85% | |
| HILL | 原始 | 63.98% | 72.15% | 76.12% | 81.00% |
| 放大 | 64.60% | 73.48% | 77.68% | 82.57% | |
| MiPOD | 原始 | 61.55% | 71.20% | 76.00% | 81.35% |
| 放大 | 62.50% | 71.73% | 77.05% | 82.63% | |
表 6 在BOSSBase数据库和0.2/0.3 bpp嵌入率下,CovNet对不同裁剪方式图像的检测准确率及其模型计算复杂度Table 6 Detection accuracy and its model computational complexity of CovNet for different cropping images under BOSSBase dataset and 0.2/0.3 bpp embedding rates |
| 尺寸 | 训练 数据 | 准确率 | 参数量/M | 计算浮点数/B | |
| 0.2 | 0.3 | ||||
| 256×256 | 原始 | 75.42% | 79.67% | 0.69 | 3.83 |
| 裁剪1 | 68.80% | 73.80% | |||
| 裁剪2 | 69.08% | 74.35% | |||
| 512×512 | 放大 | 76.55% | 81.27% | 15.32 | |
表 7 在BOSSBase数据库和0.1/0.2/0.3/0.4 bpp嵌入率下,LWENet对QGM隐写算法的检测准确率Table 7 Detection accuracy of LWENet for QGM steganography algorithms payload under BOSSBase dataset and 0.1/0.2/0.3/0.4 bpp embedding rates |
| 图像 | 准确率 | |||
| 0.1 | 0.2 | 0.3 | 0.4 | |
| 原始 | 67.17% | 75.43% | 80.93% | 84.38% |
| 放大 | 69.43% | 78.67% | 82.70% | 85.98% |
| 1 |
SIMMONS G J. The prisoners’ problem and the subliminal channel[C]//Advances in Cryptology: Proceedings of Crypto 83,1984: 51-67.
|
| 2 |
HU D H, ZHOU S N, SHEN Q, et al. Digital image steganalysis based on visual attention and deep reinforcement learning[J]. IEEE Access, 2019, 7, 25924- 25935.
|
| 3 |
DENG X Q,CHEN B L,LUO W Q,et al. Fast and effective global covariance pooling network for image steganalysis[C]//Proceedings of the ACM Workshop on Information Hiding and Multimedia Security,2019:230-234.
|
| 4 |
ZHANG S Y, ZHANG H, ZHAO X F, et al. A deep residual multi-scale convolutional network for spatial steganalysis[C]//Digital Forensics and Watermarking:17th International Workshop,2019:40-52.
|
| 5 |
ZENG J S, TAN S Q, LIU G Q, et al. WISERNet: wider separate-then-reunion network for steganalysis of color images[J]. IEEE Transactions on Information Forensics and Security, 2019, 14 (10): 2735- 2748.
|
| 6 |
WU S T, ZHONG S H, LIU Y. A novel convolutional neural network for image steganalysis with shared normalization[J]. IEEE Transactions on Multimedia, 2019, 22 (1): 256- 270.
|
| 7 |
NI D N, FENG G R, SHEN L Q, et al. Selective ensemble classification of image steganalysis via deep Q network[J]. IEEE Signal Processing Letters, 2019, 26 (7): 1065- 1069.
|
| 8 |
YOU W K, ZHANG H, ZHAO X F. A siamese CNN for image steganalysis[J]. IEEE Transactions on Information Forensics and Security, 2021, 16, 291- 306.
|
| 9 |
WENG S W, CHEN M F, YU L F, et al. Lightweight and effective deep image steganalysis network[J]. IEEE Signal Processing Letters, 2022, 29, 1888- 1892.
|
| 10 |
LUO G,WEI P,ZHU S W,et al. Image steganalysis with convolutional vision transformer[C]//2022 IEEE International Conference on Acoustics,Speech and Signal Processing (ICASSP),2022:3089-3093.
|
| 11 |
WENG S W, SUN S Y, YU L F. Fast SwT-based deep steganalysis network for arbitrary-sized images[J]. IEEE Signal Processing Letters, 2023, 30, 1782- 1786.
|
| 12 |
QIAN Y L,DONG J,WANG W,et al. Deep learning for steganalysis via convolutional neural networks[C]//Proceedings of SPIE-The International Society for Optical Engineering,2015:171-180.
|
| 13 |
XU G S, WU H Z, SHI Y Q. Structural design of convolutional neural networks for steganalysis[J]. IEEE Signal Processing Letters, 2016, 23 (5): 708- 712.
|
| 14 |
YEDROUDJ M,COMBY F,CHAUMONT M. Yedroudj-Net:an efficient CNN for spatial steganalysis[C]//2018 IEEE International Conference on Acoustics,Speech and Signal Processing,2018:2092-2096.
|
| 15 |
LI B, WEI W H, FERREIRA A, et al. ReST-Net: diverse activation modules and parallel subnets-based CNN for spatial image steganalysis[J]. IEEE Signal Processing Letters, 2018, 25 (5): 650- 654.
|
| 16 |
TAN S Q,LI B. Stacked convolutional auto-encoders for steganalysis of digital images[C]//Proceedings of Signal and Information Processing Association Annual Summit and Conference (APSIPA),2014:1-4.
|
| 17 |
YE J, NI J Q, YI Y. Deep learning hierarchical representations for image steganalysis[J]. IEEE Transactions on Information Forensics and Security, 2017, 12 (11): 2545- 2557.
|
| 18 |
ZHANG R, ZHU F, LIU J Y, et al. Depth-wise separable convolutions and multi-level pooling for an efficient spatial CNN-based steganalysis[J]. IEEE Transactions on Information Forensics and Security, 2020, 15, 1138- 1150.
|
| 19 |
BOROUMAND M, CHEN M, FRI1DRICH J. Deep residual network for steganalysis of digital images[J]. IEEE Transactions on Information Forensics and Security, 2018, 14 (5): 1181- 1193.
|
| 20 |
王晓丹, 李京泰, 宋亚飞. DDAC: 面向卷积神经网络图像隐写分析模型的特征提取方法[J]. 通信学报, 2022, 43 (5): 68- 81.
WANG X D, LI J T, SONG Y F. DDAC: a feature extraction method for convolutional neural network image steganalysis model[J]. Journal on Communications, 2022, 43 (5): 68- 81.
|
| 21 |
FRIDRICH J, KODOVSKY J. Rich models for steganalysis of digital images[J]. IEEE Transactions on Information Forensics and Security, 2012, 7 (3): 868- 882.
|
| 22 |
BUTORA J,FRIDRICH J. Extending the reverse JPEG compatibility attack to double compressed images[C]//2021 IEEE International Conference on Acoustics,Speech and Signal Processing (ICASSP),2021:2710-2714.
|
| 23 |
BUTORA J, FRIDRICH J. Reverse JPEG compatibility attack[J]. IEEE Transactions on Information Forensics and Security, 2019, 15, 1444- 1454.
|
| 24 |
FENG J L,WANG Y F,CHEN K J,et al. An effective steganalysis for robust steganography with repetitive JPEG compression[C]//2022 IEEE International Conference on Acoustics,Speech and Signal Processing (ICASSP),2022:3084-3088.
|
| 25 |
BENEŠ M,LORCH B,BÖHME R. JPEG steganalysis using leaked cover thumbnails[C]//2023 IEEE International Workshop on Information Forensics and Security (WIFS),2023:1-6.
|
| 26 |
TANG W X, TAN S Q, LI B, et al. Automatic steganographic distortion learning using a generative adversarial network[J]. IEEE Signal Processing Letters, 2017, 24 (10): 1547- 1551.
|
| 27 |
BAS P,FILLER T,PEVNÝ T. Break our steganographic system:the ins and outs of organizing BOSS[C]//International Workshop on Information Hiding,2011:59-70.
|
| 28 |
HOLUB V, FRIDRICH J, DENEMARK T. Universal distortion function for steganography in an arbitrary domain[J]. EURASIP Journal on Information Security, 2014, 2014, 1- 13.
|
| 29 |
LI B,WANG M,HUANG J W,et al. A new cost function for spatial image steganography[C]//2014 IEEE International Conference on Image Processing (ICIP),2014:4206-4210.
|
| 30 |
SEDIGHI V, COGRANNE R, FRIDRICH J. Content-adaptive steganography by minimizing statistical detectability[J]. IEEE Transactions on Information Forensics and Security, 2015, 11 (2): 221- 234.
|
| 31 |
PASZKE A,GROSS S,CHINTALA S,et al. Automatic differentiation in PyTorch[C]//31st Conference on Neural Information Processing Systems,2017: 1-4.
|
| 32 |
SELVARAJU R R,COGSWELL M,DAS A,et al. Grad-CAM:visual explanations from deep networks via gradient-based localization[C]//2017 IEEE International Conference on Computer Vision(ICCV),2017:618-626.
|
| 33 |
SHARIFZADEH M, ALORAINI M, SCHONFELD D. Adaptive batch size image merging steganography and quantized Gaussian image steganography[J]. IEEE Transactions on Information Forensics and Security, 2019, 15, 867- 879.
|
/
| 〈 |
|
〉 |