logo

SCIENCE CHINA Information Sciences, Volume 61, Issue 3: 032103(2018) https://doi.org/10.1007/s11432-016-0426-1

Countering JPEG anti-forensics based on noise level estimation

More info
  • ReceivedJun 24, 2016
  • AcceptedOct 8, 2016
  • PublishedAug 14, 2017

Abstract

Quantization artifact and blocking artifact are the two types of well-known fingerprints of JPEG compression. Most JPEG forensic techniques are focused on these fingerprints. However, recent research shows that these fingerprints can be intentionally concealed via anti-forensics, which in turn makes current JPEG forensic methods vulnerable. A typical JPEG anti-forensic method is adding anti-forensic dither to DCT transform coefficients and erasing blocking artifact to remove the trace of compression history. To deal with this challenge in JPEG forensics, in this paper, we propose a novel countering method based on the noise level estimation to identify the uncompressed images from those forged ones. The experimental results show that the proposed method achieves superior performance on several image databases with only one-dimensional feature. It is also worth emphasizing that the proposed threshold-based method has explicit physical meaning and is simple to be implemented in practice. Moreover, we analyze the strategies available to the investigator and the forger in the case of that they are aware of the existence of each other. Game theory is used to evaluate the ultimate performance when both sides adopt their Nash equilibrium strategies.


Acknowledgment

This work was supported by National Natural Science Foundation of China (Grant Nos. U1536204, 61379155, 61332012), National Natural Science Foundation of Guangdong Province (Grant No. s2013020012788), and Special Funding for Basic Scientific Research of Sun Yat-sen University (Grant No. 6177060230).


References

[1] Kwok C W, Au O C, Chui S H. Alternative anti-forensics method for contrast enhancement. In: Proceedings of the 10th International Conference on Digital-Forensics and Watermarking, Atlantic, 2011. 398--410. Google Scholar

[2] Milani S, Tagliasacchi M, Tubaro S. Antiforensics attacks to Benfords law for the detection of double compressed images. In: Proceedings of International Conference on Acoustics, Speech, and Signal Processing, Vancouver, 2013. 3053--3057. Google Scholar

[3] Qian Z X, Zhang X P. Improved anti-forensics of JPEG compression. J Syst Softw, 2014, 91: 100--108. Google Scholar

[4] Fan W, Wang K, Cayre F, et al. JPEG anti-forensics using non-parametric DCT quantization noise estimation and natural image statistics. In: Proceedings of the 1st ACM Workshop on Information Hiding and Multimedia Security, Montpellier, 2013. 117--122. Google Scholar

[5] Barni M, Tondi B. The source identification game: an information-theoretic perspective. IEEE Trans Inf Forens Secur, 2013, 8: 450--463. Google Scholar

[6] Stamm M C, Lin W S, Liu K J R. Forensics vs. anti-forensics: a decision and game theoretic framework. In: Pro- ceedings of International Conference on Acoustics, Speech, and Signal Processing, Kyoto, 2012. 1749--1752. Google Scholar

[7] Pevny T, Fridrich J. Detection of double-compression in JPEG images for applications in steganography. IEEE Trans Inf Forens Secur, 2008, 3: 247--258. Google Scholar

[8] Lukas J, Fridrich J. Estimation of primary quantization matrix in double compressed JPEG images. In: Proceedings of Digital Forensic Research Workshop, Cleveland, 2003. 1--17. Google Scholar

[9] Fu D D, Shi Y Q, Su W. A generalized Benfords law for JPEG coefficients and its applications in image forensics. In: Proceedings of SPIE 6505, Security, Steganography, and Watermarking of Multimedia Contents, San Jose, 2007. 65051L. Google Scholar

[10] He J F, Lin Z C, Wang L F, et al. Detecting doctored JPEG images via DCT coefficient analysis. In: Proceedings of 9th European Conference on Computer Vision, Graz, 2006. 423--435. Google Scholar

[11] Bianchi T, de Rosa A, Piva A. Improved DCT coefficient analysis for forgery localization in JPEG images. łinebreak In: Proceedings of International Conference on Acoustics, Speech, and Signal Processing, Prague, 2011. 2444--2447. Google Scholar

[12] Farid H. Exposing digital forgeries from JPEG ghosts. IEEE Trans Inf Forens Secur, 2009, 4: 154--160. Google Scholar

[13] Farid H. Digital Image Ballistics from JPEG Quantization. TR2006--583. 2008. Google Scholar

[14] Stamm M C, Tjoa S K, Lin W S, et al. Anti-forensics of JPEG compression. In: Proceedings of IEEE International Conference on Acoustics, Speech and Signal Processing, Dallas, 2010. 1694--1697. Google Scholar

[15] Stamm M C, Liu K J R. Anti-forensics of digital image compression. IEEE Trans Inf Forens Secur, 2011, 6: 1050--1065. Google Scholar

[16] Jiang Y W, Zeng H, Kang X G, et al. The game of countering JPEG anti-forensics based on the noise level estimation. In: Proceedings of Asian-Pacific Signal and Information Processing Association Annual Submit Conference, Taiwan, 2013. 1--9. Google Scholar

[17] Schaefer G, Stich M. UCID: an uncompressed color image database. In: Proceedings of SPIE 5307, Storage and Retrieval Methods and Applications for Multimedia, San Jose, 2004. 472--480. Google Scholar

[18] Bas P, Filler T, Pevny T. Break our steganographic system: the ins and outs of organizing BOSS. In: Proceedings of International Conference on Information Hiding, Prague, 2011. 59--70. Google Scholar

[19] Lam E Y, Goodman J W. A mathematical analysis of the DCT coefficient distributions for images. IEEE Trans Image Process, 2000, 9: 1661--1666. Google Scholar

[20] Fan Z G, de Queiroz R L. Identification of bitmap compression history: JPEG detection and quantizer estimation. IEEE Trans Image Process, 2003, 12: 230--235. Google Scholar

[21] Kirchner M, Fridrich J. On detection of median filtering in digital images. In: Proceedings of SPIE, Electronic Imaging, Media Forensics and Security II, San Jose, 2010. 1--12. Google Scholar

[22] Lai S Y, Bohme R. Countering counter-forensics: the case of JPEG compression. In: Proceedings of International Conference on Information Hiding, Prague, 2011. 285--298. Google Scholar

[23] Valenzise G, Nobile V, Tagliasacchi M, et al. Countering JPEG anti-forensics. In: Proceedings of IEEE International Conference on Image Processing, Brussels, 2011. 1949--1952. Google Scholar

[24] Valenzise G, Tagliasacchi M, Tubaro S. Revealing the traces of JPEG compression anti-forensics. IEEE Trans Inf Forens Secur, 2013, 8: 335--349. Google Scholar

[25] Rudin L I, Osher S, Fatemi E. Nonlinear total variation based noise removal algorithms. Physica D: Nonl Phenom, 1992, 60: 259--268. Google Scholar

[26] Li H D, Luo W Q, Huang J W. Countering anti-JPEG compression forensics. In: Proceedings of IEEE International Conference on Image Processing, Orlando, 2012. 241--244. Google Scholar

[27] Liu X H, Tanaka M, Okutomi M. Noise level estimation using weak textured patches of a single noisy image. łinebreak In: Proceedings of IEEE International Conference on Image Processing, Orlando, 2012. 665--668. Google Scholar

[28] Shin D H, Park R H, Yang S, et al. Block-based noise estimation using adaptive Gaussian filtering. IEEE Trans Consum Electr, 2005, 51: 218--226. Google Scholar

[29] Pyatykh S, Hesser J, Zheng L. Image noise level estimation by principal component analysis. IEEE Trans Image Process, 2013, 22: 687--699. Google Scholar

[30] Liu W, Lin W S. Additive white Gaussian noise level estimation in SVD domain for images. IEEE Trans Image Process, 2013, 22: 872--883. Google Scholar

[31] Lyu S W, Pan X Y, Zhang X. Exposing region splicing forgeries with blind local noise estimation. Int J Comput Vis, 2014, 110: 202--221. Google Scholar

[32] Osborne M J, Rubinstein A. A Course in Game Theory. Cambridge: MIT Press. 1994. Google Scholar

[33] Hespanha J P. An Introductory Course in Noncooperative Game Theory. http://www.ece.ucsb. edu/$\sim$hespanha/. Google Scholar

[34] Grant M C, Boyd S P. Graph Implementations for Non-smooth Convex Programs. Recent Advances in Learning and Control. Heidelberg: Springer-Verlag, 2008. 95--110. Google Scholar

  • Figure 1

    (a) Blocking artifact of the uncompressed Lena image; (b) the same image after JPEG compression with quality factor Q= 75; (c) the forged image.

  • Figure 2

    (Color online) Histogram of DCT coefficient of the (2, 2) subband from (a) uncompressed Lena image, (b) Lena after JPEG 75 compression, and (c) the forged image obtained after adding dither in DCT domain.

  • Figure 3

    (Color online) Histogram ($h_1(n)$ and $h_2(n)$) of neighbor pixel differences from (a) an uncompressed image, (b) the same image after JPEG 75 compression, (c) after adding anti-forensic dither, (d) after median filtering, and (e) after adding noise to obtain the final forged image.

  • Figure 4

    JPEG anti-forensics procedure.

  • Figure 5

    Noise level estimation of a test image from UCID. (a) An uncompressed image; (b) selected weak textured areas, where some overexposure areas are excluded.

  • Figure 6

    Images containing only textures. There are no enough homogeneous areas which can be selected from those images for noise level estimation.

  • Figure 7

    Compression residual of (a) an original image and (b) a forged image.

  • Figure 8

    (Color online) Noise level of the CR of an original image and a forged image under different $Q_2$.

  • Figure 9

    (Color online) Two-dimension scatter plots of features $F_1$ and $F_2$, where $F_1$ is the noise level of the test image itself and $F_2$ is the noise level of CR. The blue points denote the original images and the green stars denote the forged ones. (a) UCID; (b) BOSSbase.

  • Figure 10

    The diagram of the forgery detector. We use the metric $\rho$ [31]to measure the streaking artifact and the proposed method to measure the noise level.

  • Figure 11

    (Color online) ROC comparison in countering JPEG anti-forensics on UCID, Q= 75.

  • Figure 12

    (Color online) Nash equilibrium ROC on UCID.

  • Table 1   Detection accuracy compared with other methods on UCID (%)$^{\rm~a)}$
    Feature dimension Q = 95 Q = 85 Q = 75 Q = 65
    $F_1$ 1 99.1 99.1 99.1 99.0
    $F_2$ 1 98.6 99.0 99.1 98.9
    F 1 99.6 99.7 99.8 99.8
    Ref. [25] 100 99.4 99.6 99.7 99.7

    a) The number in bold type denotes the best performance.

  • Table 2   Detection accuracy compared with other methods on BOSSbase (%)$^{\rm~a)}$
    Feature dimension Q = 95 Q = 85 Q = 75 Q = 65
    $F_1$ 1 99.6 99.6 99.6 99.6
    $F_2$ 1 99.7 99.7 99.8 99.7
    F 1 99.9 99.9 99.9 99.9
    Ref. [25] 100 99.8 99.9 99.9 99.8

    a) The same as in Table 1.

Copyright 2020 Science China Press Co., Ltd. 《中国科学》杂志社有限责任公司 版权所有

京ICP备18024590号-1       京公网安备11010102003388号