基于多層感知卷積和通道加權的圖像隱寫檢測
doi: 10.11999/JEIT210537 cstr: 32379.14.JEIT210537
-
杭州電子科技大學通信工程學院 杭州 310018
基金項目: 國家自然科學基金(U19B2016, 60802047)
Image Steganography Detection Based on Multilayer Perceptual Convolution and Channel Weighting
-
School of Communication Engineering, Hangzhou Dianzi University, Hangzhou 310018, China
Funds: The National Natural Science Foundation of China (U19B2016, 60802047)
-
摘要: 針對目前圖像隱寫檢測模型中線性卷積層對高階特征表達能力有限，以及各通道特征圖沒有區(qū)分的問題，該文構建了一個基于多層感知卷積和通道加權的卷積神經(jīng)網(wǎng)絡(CNN)隱寫檢測模型。該模型使用多層感知卷積(Mlpconv)代替?zhèn)鹘y(tǒng)的線性卷積，增強隱寫檢測模型對高階特征的表達能力；同時引入通道加權模塊，實現(xiàn)根據(jù)全局信息對每個卷積通道賦予不同的權重，增強有用特征并抑制無用特征，增強模型提取檢測特征的質(zhì)量。實驗結果表明，該檢測模型針對不同典型隱寫算法及不同嵌入率，相比Xu-Net, Yedroudj-Net, Zhang-Net均有更高的檢測準確率，與最優(yōu)的Zhu-Net相比，準確率提高1.95%～6.15%。
-
關鍵詞:
- 隱寫檢測 /
- 卷積神經(jīng)網(wǎng)絡 /
- 多層感知卷積 /
- 通道加權
Abstract: For steganalysis, many studies have shown that convolutional neural networks have better performance than traditional artificially designed features. However, the ability of linear convolution layer to express higher-order features is limited and the feature map of each channel is not distinguished in the existing detection model which based on Convolutional Neural Networks (CNN). To solve these problems, an image steganography detection model based on Multi-layer perceptual convolution and channel weighting is constructed in this paper. The Multi-layer perceptual convolution layer (Mlpconv)is used to replace the traditional linear convolution layer to enhance the expressiveness ability of high-order features of the detection model. The channel weighting module is added to the model, which assigns different weights to each convolution channel based on global information, so that the useful features can be enhanced and the useless features can be suppressed, and the detection features extracted from the quality model can be improved. The experimental results show that the detection accuracy of proposed detection model is higher than that of Xu-Net, Yedroudj-Net, and Zhang-Net for different typical steganography algorithms and different embedding rates. And compared with the optimal Zhu-Net, the accuracy rate is increased by 1.95～6.15%. -
表 1 Yedroudj-Net[12]修改預處理層前后準確率(%)
檢測模型 WOW S-UNIWARD 0.2 bpp 0.4 bpp 0.2 bpp 0.4 bpp Yedroudj-Net[12] 72.20 85.90 63.30 77.20 改進預處理層的Yedroudj-Net 77.21 87.54 68.81 81.91 
下載: 導出CSV
表 3 通道加權前后模型的檢測準確率(%)
本文模型 WOW S-UNIWARD 0.2bpp 0.4bpp 0.2bpp 0.4bpp 通道加權前 80.72 89.59 76.94 87.66 通道加權后 81.25 90.15 77.65 88.10
下載: 導出CSV
-
[1] LIU Jia, KE Yan, ZHANG Zhuo, et al. Recent advances of image steganography with generative adversarial networks[J]. IEEE Access, 2020, 8: 60575–60597. doi: 10.1109/ACCESS.2020.2983175 [2] PEVNY T, BAS P, and FRIDRICH J. Steganalysis by subtractive pixel adjacency matrix[J]. IEEE Transactions on information Forensics and Security, 2010, 5(2): 215–224. doi: 10.1109/TIFS.2010.2045842 [3] FRIDRICH J and KODOVSKY J. Rich models for steganalysis of digital images[J]. IEEE Transactions on Information Forensics and Security, 2012, 7(3): 868–882. doi: 10.1109/tifs.2012.2190402 [4] 付章杰, 李恩露, 程旭, 等. 基于深度學習的圖像隱寫研究進展[J]. 計算機研究與發(fā)展, 2021, 58(3): 548–568. doi: 10.7544/issn1000-1239.2021.20200360FU Zhangjie, LI Enlu, CHENG Xu, et al. Recent advances in image steganography based on deep learning[J]. Computer Research and Development, 2021, 58(3): 548–568. doi: 10.7544/issn1000-1239.2021.20200360 [5] SHARIFZADEH M, ALORAINI M, and SCHONFELD D. Adaptive batch size image merging steganography and quantized Gaussian image steganography[J]. IEEE Transactions on Information Forensics and Security, 2020, 15: 867–879. doi: 10.1109/TIFS.2019.2929441 [6] LAISHRAM D and TUITHUNG T. A novel minimal distortion-based edge adaptive image steganography scheme using local complexity[J]. Multimedia Tools and Applications, 2021, 80(1): 831–854. doi: 10.1007/S11042-020-09519-9 [7] 陳君夫, 付章杰, 張衛(wèi)明, 等. 基于深度學習的圖像隱寫分析綜述[J]. 軟件學報, 2021, 32(2): 551–578. doi: 10.13328/j.cnki.jos.006135CHEN Junfu, FU Zhangjie, ZHANG Weiming, et al. Review of image steganalysis based on deep learning[J]. Journal of Software, 2021, 32(2): 551–578. doi: 10.13328/j.cnki.jos.006135 [8] TAN Shunquan and LI Bin. Stacked convolutional auto-encoders for steganalysis of digital images[C]. Signal and Information Processing Association Annual Summit and Conference (APSIPA), 2014 Asia-Pacific, Siem Reap, Cambodia, 2014: 1–4. [9] QIAN Yinlong, DONG Jing, WANG Wei, et al. Deep learning for steganalysis via convolutional neural networks[J]. SPIE, 2015, 9409. [10] XU Guanshuo, WU Hanzhou, and SHI Yunqing. Structural design of convolutional neural networks for steganalysis[J]. IEEE Signal Processing Letters, 2016, 23(5): 708–712. doi: 10.1109/LSP.2016.2548421 [11] YE Jian, NI Jiangqun, and YI Yang. Deep learning hierarchical representations for image steganalysis[J]. IEEE Transactions on Information Forensics and Security, 2017, 12(11): 2545–2557. doi: 10.1109/TIFS.2017.2710946 [12] YEDROUDJ M, COMBY F, and CHAUMONT M. Yedrouj-Net: An efficient CNN for spatial steganalysis[C]. 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Calgary, Canada, 2018: 2092–2096. [13] ZHANG Ru, ZHU Feng, LIU Jianyi, 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. doi: 10.1109/TIFS.2019.2936913 [14] HOLUB V and FRIDRICH J. Designing steganographic distortion using directional filters[C]. 2012 IEEE International Workshop on Information Forensics and Security (WIFS), Costa Adeje, Spain, 2012: 234–239. [15] HOLUB V, FRIDRICH J, and DENEMARK T. Universal distortion function for steganography in an arbitrary domain[J]. EURASIP Journal on Information Security, 2014, 2014: 1. doi: 10.1186/1687-417X-2014-1 [16] MEMISEVIC R, ZACH C, HINTON G E, et al. Gated softmax classification[C]. The 23rd International Conference on Neural Information Processing Systems, Red Hook, USA, 2010: 1603–1611. [17] LIN Min, CHEN Qiang, and YAN Shuicheng. Network in network[Z]. ArXiv: 1312.4400, 2013. [18] HU JIE, SHEN Li, ALBANIE S, et al. Squeeze-and-excitation networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 42(8): 2011–2023. doi: 10.1109/TPAMI.2019.2913372 [19] BAS P, FILLER T, and TOMÁS PEVNÝ T. “Break Our Steganographic System”: The ins and outs of organizing BOSS[C]. Information Hiding 13th International Conference, Prague, Czech Republic, 2011: 59–70. [20] GLOROT X and BENGIO Y. Understanding the difficulty of training deep feedforward neural networks[J]. Journal of Machine Learning Research, 2010, 9: 249–256. -
圖(7) / 表(3)
計量
- 文章訪問數(shù): 939
- HTML全文瀏覽量: 734
- PDF下載量: 92
- 被引次數(shù): 0
下載:
