FTM1628 Adaptive Steganalysis Based on Embedding Probabilities of Pixels – IEEE MATLAB Project 2016 – 2017


FTM1628 Adaptive Steganalysis Based on Embedding Probabilities of Pixels – IEEE MATLAB Project 2016 – 2017


In modern steganography, embedding modifications are highly concentrated on the textural regions within an image, as such regions are difficult to model for steganalysis. Previous studies have shown that compared with non-adaptive strategies, this content adaptive strategy achieves stronger security against existing steganalysis. Based on the experiments and analyses, however, we found that this embedding property would inevitably lead to a large limitation in existing adaptive steganography. That is, it is possible for steganalyzers to estimate the regions that have probably been modified after data hiding. In this paper, we propose an adaptive steganalytic scheme based on embedding probabilities of pixels. The main idea of our scheme is that we assign different weights to different pixels in feature extraction. For those pixels with high embedding probabilities, their corresponding weights are larger since they should contribute more to steganalysis, and vice versa. By doing so, we can concentrate our attention on the regions that have probably been modified and significantly reduce the impact of other unchanged smooth regions. It is expected that our proposed method is an improvement on the existing steganalytic methods, which usually assume every pixel has the same contribution to steganalysis. The extensive experiments evaluated on four typical adaptive steganographic methods have shown the effectiveness of the proposed scheme, especially for low embedding rates, for example, lower than 0.20 bpp.

Image steganography is the art and science of concealing a secret message in an image by modifying image pixels and/or frequency coefficients. The most important requirement in steganography is undetectability. Therefore, various steganographic methods attempt to embed messages in an imperceptible manner so that the resulting stego is similar to its corresponding cover image visually and statistically. LSB (least significant bit) replacement is the simplest steganographic method. However, it introduces some asymmetry artifacts into stegos, and thus it is easily detected using some steganalytic methods such as Chi-squared attack, regular/singular group analysis and sample pair analysis.

The existing steganalytic methods, such as SPAM, SRM, PSRM, and LBP (local binary pattern)-based method, can be regarded as non-adaptive steganalysis, which means that they deal with every pixel within an image equally in feature extraction. Based on the analysis however, we found that the probability map and the modification map are highly correlative. Therefore, we should pay more attention to the pixels with high embedding probabilities when extracting the steganalytic features. This is the key idea of adaptive steganalysis. Recently, several related works of adaptive steganalysis have been proposed. In the idea of adaptive steganalysis was firstly proposed based on game theory model. The first algorithm that implements the idea was the specialized WS (weighted stego-image) steganalysis against naïve adaptive LSB replacement steganography.


  1. Obtaining Embedding Probabilities via Optimal Simulator
  2. Performance Analysis with Correlation Test
  3. Robustness Analysis against Noise Contamination
  4. Additive noise
  5. Report Generation

Obtaining Embedding Probabilities via Optimal Simulator:

In the case where the adaptive steganographic method is designed under the framework of minimizing the additive distortion function and the embedding method is known, embedding probabilities can be accurately estimated by the optimal simulator. For comparative studies, the original nonadaptive steganalysis SRM, our previous work, the method and the proposed method are included in the experiments. Three adaptive methods, WOW, HUGO BD, S-UNIWARD, designed under the same framework as that illustrated in evaluated. For each steganography, six different embedding rates ranging from 0.05 bpp to 0.5 bpp are tested. Please note that in our previous method, the performance is affected by a parameter P, and the detection error with the best parameter P is given in all experiments.

Our proposed method and the method are compared under the same steganalytic features. From the corresponding values in the 4th and the 6th column of each table, it can be seen that the detection performance between the two methods is very close using the SRM features. Similar results can be obtained via comparing the values in the 5th and the 7th column using the SRMd2 features. Experimental results indicate that in the case that embedding method is known, the proposed method can achieve equally good results with the method.

Performance Analysis with Correlation Test:

Based on the above experiments and analyses, it is noted that how the locations of the modified pixels are estimated plays an important role in the proposed adaptive steganalysis. The modifications after adaptive steganography are largely located in the textural regions, so that it is possible to estimate these modifications more accurately for steganalysis. That is why the proposed strategy improves the detection performance in adaptive steganography. Please note that the proposed strategy cannot improve the detection performance for non-adaptive steganographic methods, as the embedding probabilities of all pixels are exactly the same in this case, and the correlation coefficients are usually close to zero

Robustness Analysis against Noise Contamination:

To evaluate the performance of detecting adaptive steganography for a given embedding rate, in the training stage, we calculate embedding probabilities of 5,000 randomly selected cover images and their corresponding stegos respectively according to the embedding rate, and then we obtained their adaptive SRM features as described. The resulting features are then used to train an ensemble classifier. In the testing stage, we perform the same operations for any testing image to obtain adaptive feature, which is then fed into the corresponding classifier, as it was in most existing steganalytic literature, without considering the unknown embedding payload problems. The detection performance is quantified using the testing error, which is the average of the false alarm rate and the missed detection rate.

Additive noise:

Based on the above simulation experiments, we obtain the three following observations. The multiplicative noise barely affects the detection performance of the proposed adaptive steganalysis against the three steganographic methods; For the spatial jitter and the additive noise, the detection errors would increase with increasing the strength of noise, especially when the embedding rate is low, e.g., 0.05 bpp. However, for high embedding rates, such as 0.3 bpp and 0.5 bpp, the increases of detection errors are relatively smaller compared to the corresponding case of 0.05 bpp; The increase of detection errors for S-UNIWARD are relatively smaller compared to the results for WOW and HUGO BD under the same embedding rate, the same type of noise and the same noise strength. The performance of the adaptive steganalysis may be poorer than non-adaptive steganalysis when the probabilities are corrupted by spatial jitter or additive noise with high noise strength.

Report Generation:

Based on our extensive experiments and analyses, we have found that the embedding probabilities of pixels within an image are not uniform for adaptive steganography, and that most pixels located in textural regions usually have much higher probabilities than those located in smooth regions. Based on this embedding property, therefore, we have proposed an adaptive steganalytic strategy using embedding probabilities estimated via either optimal simulator and/or re-embedding random experiments. The experimental results evaluated with four typical adaptive steganographic methods WOW, SUNIWARD, HUGO BD, and EA, have shown the effectiveness of the proposed strategy. The main contributions of this paper are as follows. We have analysed the common limitation of existing adaptive steganography, and find out that embedding probabilities can provide us with useful information for improving classical non-adaptive steganalysis. We have introduced a new steganalytic strategy for existing adaptive steganography via adjusting the contributions of pixels to steganalysis according to the estimated embedding probabilities, and combine some modern steganalytic features, i:e:, SRM, in the proposed strategy.


HARDWARE REQUIREMENTS:                               

  • System : Intel Core i3 2.8 GHz.
  • Hard Disk : 250 GB.
  • Monitor : 15” VGA Colour.
  • Mouse : Logitech.
  • Ram : 1 GB.


  • Operating system      :  Windows 7.
  • Software                      : Matlab with Simulink


[1] A.Westfeld and A. Pfitzmann. Attacks on steganographic systems. In Proc. 3rd International Workshop on Information Hiding, volume 1768 of Lecture Notes in Computer Science, pages 61–75. Dresden, Germany, September 1999.

[2] P. Bas, T. Filler, and T. Pevn´y. ‘Break our steganographic system’: The ins and outs of organizing BOSS. In Proc. 13th International Workshop on Information Hiding, volume 6958 of Lecture Notes in Computer Science, pages 59–70. Prague, Czech Republic, May 2011.

[3] D.C.Wu and W.H.Tsai. A steganographic method for images by pixel value differencing. Pattern Recognition Letters, 24:1613–1626, June 2003.

[4] D.Zou, Y.Q.Shi, W.Su, and G.Xuan. Steganalysis based on markov model of threshold prediction-error image. In Proc. 2006 IEEE International Conference on Multimedia and Expo, pages 1365–1368, Toronto, Canada, July 2006.

[5] J. Fridrich and T. Filler. Practical methods for minimizing embedding impact in steganography. In Proc. SPIE, Electronic Imaging, Security, Steganography, and Watermarking of Multimedia Contents IX, volume 6505, pages 650502–1–650502–15, January 2007.