Your browser does not support JavaScript!
http://iet.metastore.ingenta.com
1887

access icon openaccess Application of quantisation-based deep-learning model compression in JPEG image steganalysis

This is an open access article published by the IET under the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0/)
Received 18/07/2018, Accepted 26/07/2018, Published 16/08/2018

Steganography can hide secret information in an innocent cover medium. Its opponent is steganalysis, which is used to discriminate whether a suspicious carrier contains a hidden message or not. With the rapid development of deep-learning frameworks, deep-learning-based steganalytic models have hold the dominant position in the field of steganalysis. In recent years, some scholars have successfully utilised model compression methods in the field of image classification. However, as far as the authors know, no prior works are devoted to the application of model compression methods in the field of deep-learning-based steganalysis. In this study, the authors explore the effect of two quantisation schemes, namely 8-bit calculation and floating-point calculation, on the performance of XuNet, a state-of-the-art deep-learning steganalytic model. The experimental results show that the two deep-learning model quantisation schemes are applicable to steganalysis. It is even possible to compress the network size while retaining satisfactory performance.

Inspec keywords: image coding; data compression; learning (artificial intelligence); floating point arithmetic; quantisation (signal); steganography

Other keywords: deep learning based steganalytic models; 8-bit calculation; cover medium; XuNet; hidden message; floating point calculation; steganography; JPEG image steganalysis; secret information hiding; suspicious carrier; quantisation based deep learning model compression; image classification

Subjects: Data security; Computer vision and image processing techniques; Knowledge engineering techniques; Cryptography; Image and video coding

References

    1. 1)
      • 2. Xu, G.: ‘Deep convolutional neural network to detect J-UNIWARD’. Proc. 5th ACM Information Hiding and Multimedia Security Workshop (IH&MMSec'2017), Philadelphia, USA, June 2017, pp. 67–73.
    2. 2)
      • 15. Srivastava, N., Hinton, G., Krizhevsky, A., et al: ‘Dropout: a simple way to prevent neural networks from overfitting’, J. Mach. Learn. Res., 2014, 15, pp. 1929–1958.
    3. 3)
      • 11. Lin, Z., Courbariaux, M., Memisevic, R., et al: ‘Neural networks with few multiplications’, arXiv:1510.03009v3 [cs.LG], 26 February 2016.
    4. 4)
      • 5. Zeng, J., Tan, S., Li, B., et al: ‘Large-scale JPEG steganalysis using hybrid deep-learning framework’, IEEE Trans. Inf. Forensics Sec., 2018, 13, (5), pp. 1242–1257.
    5. 5)
      • 1. Holub, V., Fridrich, J., Denemark, T.: ‘Universal distortion function for steganography in an arbitrary domain’, EURASIP J. Inf. Secur., 2014, 2014, (1), pp. 1–13.
    6. 6)
      • 8. Han, S., Pool, J., Tran, J., et al: ‘Learning both weights and connections for efficient neural networks’. Advances in Neural Information Processing Systems 28 (NIPS 2015), Montreal, Canada, December 2015.
    7. 7)
      • 12. Kim, Y.-D., Park, E., Yoo, S., et al: ‘Compression of deep convolutional neuralnetworks for fast and low power mobile applications’, arXiv:1511.06530v2 [cs.CV], 24 February 2016.
    8. 8)
      • 10. Wen, W., Wu, C., Wang, Y., et al: ‘Learning structured sparsity in deep neural networks’. 30th Conf. on Neural Information Processing Systems (NIPS 2016), Barcelona, Spain, December 2016.
    9. 9)
      • 7. Iandola, F.N.: ‘Squeezenet: alexnet-level accuracy with 50X fewer parameters and <0.5 MB model size’, arXiv preprint arXiv:1602.07360, 2016.
    10. 10)
      • 9. Han, S., Mao, H., Dally, W.J.: ‘Deep compression: compressing deep neural network with pruning, trained quantization and huffman coding’, arXiv preprint arXiv:1510.00149, 2015.
    11. 11)
      • 6. He, K., Zhang, X., Ren, S., et al: ‘Deep residual learning for image recognition’. Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR’ 2016), Las Vegas, USA, June 2016, pp. 770–778.
    12. 12)
      • 14. Molchanov, D., Ashukha, A., Vetrov, D.: ‘Variational dropout sparsifies deep neural networks’, arXiv:1701.05369v3 [stat.ML], 13 June 2017.
    13. 13)
      • 4. Zeng, J., Tan, S., Li, B.: ‘Pre-training via fitting deep neural network to rich-model features extraction procedure and its effect on deep learning for steganalysis’. Proc. Media Watermarking, Security, and Forensics, Part of IS&T Int. Symp. on Electronic Imaging (EI'2017), Burlingame, USA, February 2017, pp. 44–49.
    14. 14)
      • 13. Judd, P., Albericio, J., Hetherington, T., et al: ‘Reduced-precision strategies for boundedmemory in deep neural nets’, arXiv:1511.05236v4 [cs.LG], 8 January 2016.
    15. 15)
      • 3. Xu, G., Wu, H.Z., Shi, Y.Q.: ‘Structural design of convolutional neural networks for steganalysis’, IEEE Signal Process. Lett., 2016, 23, (5), pp. 708–712.
http://iet.metastore.ingenta.com/content/journals/10.1049/joe.2018.8299
Loading

Related content

content/journals/10.1049/joe.2018.8299
pub_keyword,iet_inspecKeyword,pub_concept
6
6
Loading
Print this page
This is a required field
Please enter a valid email address