VRL an eye

Resisting Blind Steganalysis

People:

Anindya Sarkar, Kaushal Solanki, B. S. Manjunath

Objective:

A simple approach for active image steganography is proposed that can successfully resist recent blind steganalysis methods, in addition to surviving JPEG compression attacks.

Methodology

The scheme is titled "Yet Another Steganographic Scheme" (YASS). It is based on embedding data in randomized locations so as to defeat the self-calibration process, commonly used by blind steganalysis schemes. The self-calibration based steganalysis schemes look at recovering an accurate estimate of the cover statsistics by cropping a few pixels from the stego image and then estimating first and second order feature statistics. By hiding in different 8x8 blocks,  which are chosen randomly in a BxB block (where B is called the big block size and is greater than 8), the block structure seen by a steganalyst  becomes desynchronized with the actual 8x8 blocks used for hiding and thus, we are successful in confusing the detector while it tries to estimate the cover image statistics.

The penalty we pay for using a non-standard (irregular) 8x8 grid is that  the errors introduced by JPEG compression (which considers regular 8x8 grids) are higher than if we hide in regular 8x8 blocks. For lossless data recovery, we have to use a higher redundancy factor in our error correction framework.

Method details:

Here, we deal only with luminance images. If the training or testing set contains color images, then the stego image will be a Y-channel image only with embedded data. In the absence of hiding, a luminance image with no embedded data will be considered as the cover. Another aspect of our embedding code is that we do not specify a certain bit budget- we let the scheme hide as much data as it can. Thus, we assume that the embedded bits are already error-correction coded. The repeat-accumulate coding based error-correction procedure used by our method is described in detail in [7]. Once we run our error correcting framework, we can figure out the redundancy factor needed to successfully retrieve the data.

Detection Results:

The steganographic security of our scheme is evaluated against the following blind steganalysis schemes. The names in bold are the ones used to denote the steganalysis schemes.

  1. Farid: 72-dimensional feature vector based on moments in the wavelet domain [1].
  2. PF-23: Pevny and Fridrich’s 23-dimensional DCT feature vector [2].
  3. PF-274: Pevny and Fridrich’s 274-dimensional feature vector that merges Markov and DCT features [3].
  4. DCT hist.: Histogram of DCT coefficients from a low-frequency band [4].
  5. Xuan-39: Spatial domain steganalysis proposed by Xuan et al [5] (a 39-dimensional feature vector).
  6. Chen-324: JPEG steganalysis based on statistical moments of wavelet characteristic functions proposed by Chen et al [6].

Table 1:  JPEG dataset: Steganalysis results for randomized block based hiding, when big-block size B is varied. It can be seen that the detection is random for most of the configurations

QFh  (used for hiding)
QFa  (used for steganalysis)
Steganalysis method
Detection accuracy: B=9
Detection accuracy: B=14
50
50
Farid
0.52
0.51
50
75
Farid
0.55
0.51
75
75
Farid
0.52
0.51
50
50
PF-23
0.56
0.54
50
75
PF-23 0.59
0.60
75
75
PF-23 0.53
0.52
50
50
PF-274
0.58
0.55
50
75
PF-274 0.77
0.65
75
75
PF-274 0.59
0.54
50 50 DCT-hist
0.53
0.53
50 75 DCT-hist 0.64
0.54
75 75 DCT-hist 0.55
0.53
50 50 Xuan-39
0.54
0.51
50 75 Xuan-39 0.63
0.53
75 75 Xuan-39 0.52
0.52
50 50 Chen-324
0.57
0.54
50 75 Chen-324 0.75
0.55
75 75 Chen-324 0.54
0.53

Table 2: Steganalysis results, in terms of detection accuracy, for comparing the randomized block based scheme (YASS) with OutGuess and Steghide schemes, used at rates of 1/10 for the TIFF image dataset. For OutGuess and Steghide, the images are JPEG compressed using a quality factor of QFa before being presented to the steganographic scheme. Note that the QFh parameter is applicable only for the YASS scheme.

QFh (used for hiding)
QFa (used for steganalysis)
Steganalysis method
YASS
OutGuess - 1/10
Steghide - 1/10
50
50
Farid 0.51
0.74
0.50
50
75
Farid 0.53
0.59
0.50
75
75
Farid 0.50
0.59
0.50
50
50
PF-23 0.53
0.98
0.78
50
75
PF-23 0.59
1.00
0.99
75
75
PF-23 0.54
1.00
0.99
50
50
PF-274 0.52
1.00
0.98
50
75
PF-274 0.72
1.00 1.00
75
75
PF-274 0.56
1.00 1.00
50
50
DCT-hist 0.51
0.95
0.59
50
75
DCT-hist 0.60
1.00 0.91
75
75
DCT-hist 0.52
1.00 0.91

Acknowledgements

This research is supported by a grant from ONR #N00014-05-1-0816.

Publications:

  1. Kaushal Solanki, Anindya Sarkar and B. S. Manjunath,
    "YASS: Yet Another Steganographic Scheme that Resists Blind Steganalysis"
    9th International Workshop on Information Hiding, Saint Malo, Brittany, France, Jun. 2007.
    [abstract] [PDF] [BibTex]

    Abstract preview: "A new, simple, approach for active steganography is pro- posed in this paper that can successfully resist recent blind steganaly- sis methods, in addition to surviving distortion constrained attacks. ..." [more]

  2. Anindya Sarkar, Kaushal Solanki and B. S. Manjunath,
    "Further Study on YASS: Steganography Based on Randomized Embedding to Resist Blind Steganalysis"
    Proc. SPIE - Security, Steganography, and Watermarking of Multimedia Contents (X), San Jose, California, Jan. 2008.
    [abstract] [PDF] [BibTex]

    Abstract preview: "We present further extensions of yet another steganographic scheme (YASS), a method based on embedding data in randomized locations so as to resist blind steganalysis. YASS is a JPEG steganographic te..." [more]

References:

[1]: Farid, H.: http://www.cs.dartmouth.edu/farid/research/steg.m. (Code for generating wavelet-based feature vectors for steganalysis.)
[2]: Pevny, T., Fridrich, J.: Multi-class blind steganalysis for JPEG images. In: Proc. of SPIE, San Jose, CA (2006)
[3]: Pevny, T., Fridrich, J.: Merging Markov and DCT features for multi-class JPEG steganalysis. In: Proc. of SPIE, San Jose, CA (2007)
[4]: Solanki, K., Sullivan, K., Madhow, U., Manjunath, B.S., Chandrasekaran, S.: Provably secure steganography: Achieving zero K-L divergence using statistical restoration. In: Proc. ICIP. (2006) 125–128
[5]: Xuan, G., Shi, Y.Q., Gao, J., Zou, D., Yang, C., Yang, C., Zhang, Z., Chai, P., Chen, C., Chen, W.: Steganalysis based on multiple features formed by statistical moments of wavelet characteristic functions. In: Lecture notes in computer science: 7th International Workshop on Information Hiding. (2005)
[6]: Chen, C., Shi, Y.Q., Chen, W., Xuan, G.: Statistical moments based universal steganalysis using JPEG-2D array and 2-D characteristic function. In: Proc. ICIP, Atlanta, GA, USA (2006) 105–108
[7]: Solanki, K., Jacobsen, N., Madhow, U., Manjunath, B.S., Chandrasekaran, S.: Robust image-adaptive data hiding based on erasure and error correction. IEEE Trans. on Image Processing 13(12) (2004) 1627 –1639