Steganalysis of a PVD-based content adaptive image steganography

doi:10.1016/j.sigpro.2013.03.029

Signal Processing

Volume 93, Issue 9, September 2013, Pages 2529-2538

https://doi.org/10.1016/j.sigpro.2013.03.029 Get rights and content

Abstract

Pixel-value-differencing (PVD) is a well-known technique for content adaptive steganography. By this technique, secret data are embedded into the differences of adjacent pixels. Recently, a new PVD-based steganographic method is proposed by Luo et al. Besides realizing adaptive embedding using PVD, the new method also exploits a pairwise modification mechanism to reduce the distortion. In this work, a targeted detector is devised to detect the new PVD-based steganography. We show that although content adaptive approach may enhance the stego-security, Luo et al.'s PVD-based scheme is not a good choice for realizing adaptive embedding since it contains a serious design flaw in data embedding procedure and this flaw can lead to possible attacks. More specifically, by counting the differences of adjacent pixels in both vertical and horizontal directions, a folded difference-histogram is generated and we show that Luo et al.'s PVD-based method may arise significant artifact to this histogram which can be exploited for reliable detection. Experimental results verify that Luo et al.'s PVD-based method can be detected by the proposed detector even at a low embedding rate of 0.05 bits per pixel.

Highlights

•
PVD is a well-known technique for content adaptive steganography.
•
A targeted detector is devised to detect a recently proposed PVD-based method.
•
The PVD-based method can be detected even at a low ER of 0.05 bpp.
•
A theoretical analysis of the proposed method is provided.

Introduction

Steganalysis algorithms can be generally classified into two categories: targeted and universal [1], [2], [3], [4]. Targeted algorithms aim to identify the existence of hidden data embedded by a specific steganographic method, whereas universal algorithms intend to detect a wide range of steganography. We consider digital image as cover data and study the technique of targeted steganalysis in this work.

It is widely accepted that taking the characteristics of natural image into account may enhance stego-security. For example, it is obvious that embedding modifications operated in rough regions of a natural image are less perceptible than that in flat regions. Besides, the slight modifications to rough regions cannot be easily perceived by analyzing usual image statistics since the embedding noise is covered by the inherent noise. Thus the content adaptive approach for steganography has the potential to provide a higher level of security. Based on this consideration, Wu et al. proposed the so-called pixel-value-differencing (PVD) steganography [5], in which the difference value of a pixel pair is considered as a smoothness measurement and more data bits will be embedded into the pair if its difference is relatively large. Thereafter, numerous PVD-based methods are proposed [6], [7], [8], [9] and their security are also discussed [10], [11], [12], [13].

Recently, a new PVD-based method is proposed by Luo et al. [14]. By incorporating PVD with the pairwise embedding algorithm of Mielikainen [15], this method can realize content adaptive embedding and meanwhile provide a better PSNR compared with some previously proposed PVD-based methods. The experimental results reported in [14] show that this method is secure in resisting state-of-the-art steganalyzers.

In this work, we propose a targeted detector to detect Luo et al.'s PVD-based method. We show that although content adaptive embedding is a way to enhance stego-security, Luo et al.'s PVD-based scheme is not a good choice for realizing adaptive embedding since it contains a serious design flaw in data embedding procedure and this flaw can lead to possible attacks. More specifically, by counting the differences of adjacent pixels in both vertical and horizontal directions, a folded difference-histogram is generated and we show that Luo et al.'s PVD-based method may arise significant artifact to this histogram which can be exploited for reliable detection. By our detector, Luo et al.'s PVD-based method can be detected even at a low embedding rate (secret data bits embedded per pixel, ER for short) of 0.05 bits per pixel (bpp).

The rest of this paper is organized as follows. First, the embedding procedure of Luo et al.'s method is described in Section 2. Then, to better present our idea, we consider to detect a simple case of Luo et al.'s method in Section 3.1. A theoretical analysis of our method for this simple case is also provided. Next, the proposed detector for the general case of Luo et al.'s method is introduced in Section 3.2. Finally, experimental results are reported and conclusions are drawn in 4 Experimental results, 5 Conclusion, respectively.

Section snippets

Embedding procedure of Luo et al.'s method

The data embedding procedure of Luo et al.'s method is described step by step as follows. Some remarks are also included in the description.

Step 1 (pixel pair partition): First, for a pre-selected integer $Bz \in {1, 4, 8, 12}$ , divide the cover image into non-overlapped blocks of $Bz \times Bz$ pixels. Then, for each pixel block, rotate it by a pseudo-random degree chosen from {0°,90°,180°,270°}. Next, rearrange the resulting image as a row vector V by raster scanning. Finally, divide V into non-overlapped

A case study with the block size Bz=1

We consider in this subsection a simple case of Luo et al.'s method where the block size Bz is fixed at 1. Such a case equivalently skips the processing before image raster scanning in Step 1 of Section 2.

Define the difference-histogram of cover image as $h_{c} (t) = \frac{| {(x, y) \in V : x - y = t} |}{| V |}$ where V is the set defined in Step 1 of Section 2. The corresponding difference-histogram of stego image noted as h_s can be defined in a similar way. From Table 1, the change of difference-histogram due to data

Experimental results

The experiments are performed on the BOSSBase v1.00 database [19] which contains 10,000 512×512 gray-scale images. The stego images are obtained by Luo et al.'s method with random $Bz \in {1, 4, 8, 12}$ and different ER. Then the detector D defined in (22) is computed for both cover and stego images.

The proposed detector is evaluated by comparing it with following prior arts: the second order SPAM proposed by Pevny et al. [20], and the recently proposed detector [12] of Tan and Li which is specifically

Conclusion

In this work, we devised a targeted detector for detecting the recently proposed PVD-based embedding method [14]. We have shown that this PVD-based method may arise significant artifact to the difference-histogram, and it can be well detected even for an ER as low as 0.05 bpp. In this light, we conclude that the PVD-based method [14] is not a good choice to realize content adaptive embedding.

Acknowledgments

The authors would like to thank Dr. Shunquan Tan of Shenzhen University, Shenzhen, China, for providing us the source code in [12]. The work of Bin Li was supported by National Natural Science Foundation of China (61103174) and Fundamental Research General Program of Shenzhen City (JCYJ20120613113535357). The work of Xiangyang Luo was supported by Postdoctoral Science Foundation of China (2012T50842).

References (20)

X. Luo et al.
A review on blind detection for image steganography
Signal Processing
(2008)
A. Cheddad et al.
Digital image steganographysurvey and analysis of current methods
Signal Processing
(2010)
D.C. Wu et al.
A steganographic method for images by pixel-value differencing
Pattern Recognition Letters
(2003)
X. Zhang et al.
Vulnerability of pixel-value differencing steganography to histogram analysis and modification for enhanced security
Pattern Recognition Letters
(2004)
J. Fridrich
Steganography in Digital MediaPrinciples, Algorithms, and Applications
(2010)
B. Li et al.
A survey on image steganography and steganalysis
Journal of Information Hiding and Multimedia Signal Processing
(2011)
H.C. Wu et al.
Image steganographic scheme based on pixel-value differencing and LSB replacement methods
IEE Proceedings Vision, Image & Signal Processing
(2005)
K. Hempstalk, Hiding behind corners: using edges in images for better steganography, in: Proceedings of the Computing...
C.H. Yang et al.
Adaptive data hiding in edge areas of images with spatial LSB domain systems
IEEE Transactions on Information Forensics and Security
(2008)
C.H. Yang, S.J. Wang, C.Y. Weng, Analyses of pixel-value-differencing schemes with LSB replacement in stegonagraphy,...

There are more references available in the full text version of this article.

Cited by (18)

Robustness enhancement against adversarial steganography via steganalyzer outputs
2022, Journal of Information Security and Applications
Citation Excerpt :
Since syndrome-trellis codes (STCs) [10,11] perform near the maximum theoretical bound at the second task, steganography research mostly focuses on the design of the cost function, such as WOW [12], UNIWARD [13], HILL [14], MiPOD [15], UERD [16] and J-MiPOD [17] etc. With the development of steganography, many steganalysis methods [18–24] have been proposed. Steganalysis is an image binary classification task that aims to classify cover and stego images.
Recently, CNN (convolutional neural network) steganalyzers have significantly outperformed handcrafted features in detecting steganography. However, adversarial steganography has challenged the applications of them in the real world. Adversarial steganography can easily deceive the target CNN steganalyzer while sending secret messages. In this paper, a general framework is proposed. It that can improve the robustness of CNN steganalyzers against adversarial steganography while keeping detecting cover and conventional stego images accurately. Specifically, a rough filter that filters adversarial stego images out of the input data is set. It exploits the differences between cover and adversarial stego images on probabilistic outputs of the target CNN steganalyzer and a handcrafted steganalyzer. Extensive experiments show that the proposed framework can significantly improve the robustness of CNN steganalyzers. In the real-world scenario where cover, conventional stego and adversarial stego images are mixed, the robustness enhanced CNN steganalyzers can achieve the optimal overall performance.
Reversal of pixel rotation: A reversible data hiding system towards cybersecurity in encrypted images
2022, Journal of Visual Communication and Image Representation
Citation Excerpt :
The statistical characteristics of them are extremely similar so that the PVD statistical characteristic may be revealed. After embedding, the PVD distribution has a higher consistency which can also resist the PVD steganalysis [44]. Therefore, we can conclude that our proposed schemes have a dependable capability to guarantee security.
Due to privacy and security concerns, the researches of reversible data hiding in encrypted images (RDHEI) have become increasingly important. Conventional schemes vacate the spare room after image encryption (VRAE) suffer from the low embedding rate, high error rate of data extraction, and imperfect image recovery. To address these issues, we propose a separable reversible data hiding scheme for encrypted images that utilizes a novel pixel rotation technique to embed data into fully encrypted images. The block complexities of four decrypted rotation states are considered when recovering image. To realize perfect image recovery, we further devise a lossless version (LPR-RDHEI). Experimental results demonstrate that the proposed PR-RDHEI scheme achieves an embedding rate of 0.4994 bpp on average and ensures lossless data extraction. Meanwhile, the proposed LPR-RDHEI scheme still has a 0.4494 bpp embedding rate on average. The embedding rates of our two schemes are significantly improved compared with state-of-the-arts.
High-capacity adaptive steganography based on LSB and Hamming code
2020, Optik
Citation Excerpt :
Therefore, Luo et al. [6] pointed out that the statistical characteristics of the edge regions are more complex than the statistical features of the smooth regions and will be better preserved after data hiding. They proposed a new adaptive steganography that received a lot of attention [7,8]. The adaptive steganography technique [6,9,10] first obtains a statistical global feature of the cover image indicating where the change was made before embedding the LSB or DCT coefficients.
The development of the Internet provides the ability to transfer large amounts of data quickly and easily. However, sensitive data is easily intercepted by hackers on the Internet. Image steganography ensures the secure transmission of information. In order to improve the capacity of hidden secret data and provide imperceptible hidden image quality, this paper proposes a hybrid steganography method based on least significant bit (LSB) replacement and Hamming code (HLAH). Information security is also increased by using two different steganographic methods. Since the sharp areas of the image can tolerate more changes than the smooth areas, more secret messages are embedded in the edge regions of the image and a small amount of information are embedded in the smooth regions. Finally, the performance of the proposed HLAH method is evaluated by measuring payload capacity, mean square error (MSE), peak signal-to-noise ratio (PSNR), and histogram analysis. The experimental results show that the proposed HLAH method not only has greater embedding capacity than other existing methods, but also ensures higher image quality.
A novel image steganography technique based on quantum substitution boxes
2019, Optics and Laser Technology
Citation Excerpt :
The key benefit of this technique is obtaining high embedding capacity. However, LSB-based mechanisms suffer from various attacks and steganalysis as shown in [21–28]. To avoid this deficiency, the secret object should be encrypted before embedding it into the cover image.
Substitution boxes play an essential role in designing secure cryptosystems. With the evolution of quantum technologies, current data security mechanisms may be broken due to their construction based on mathematical computation. Quantum walks, a universal quantum computational model, play an essential role in designing quantum algorithms. We utilize the benefits of quantum walks to present a novel technique for constructing substitution boxes (S-boxes) based on quantum walks (QWs). The performance of the presented QWs S-box technique is evaluated by S-box evaluation criteria, and our results prove that the constructed S-box has vital qualities for viable applications in security purposes. Furthermore, a new technique for image steganography is constructed. The proposed technique is an integrated mechanism between classical data hiding and quantum walks to achieve better security for the embedded data. The embedding and extraction procedures are controlled by QWs S-box. The inclusion of cryptographic QWs S-box ensures the security of both embedding and extraction phases. At the extraction phase, only the stego image and the secret values are needed for constructing the secret data. Experimental results demonstrate that the presented technique has a high-security, high embedding capacity and good visual quality.
Matrix embedding in finite abelian group
2015, Signal Processing
Citation Excerpt :
Based on this consideration, Wu et al. proposed the so-called pixel-value-differencing (PVD) steganography [24], in which the difference value of a pixel pair is considered as the smoothness measurement and more data bits will be embedded into a pair if its difference is relatively larger. To the best of our knowledge, the PVD-based approach is the first attempt to realize content-adaptive steganography, and it is extensively studied in the literature including both performance enhancement and security discussions [25–32]. Later on, Fridrich et al. proposed the wet paper code (WPC) based steganography [33,34].
Matrix embedding (ME) is a well-known steganographic scheme that can improve the embedding efficiency of steganography. In ME, the sender and recipient agree on a matrix in advance, and the message will be embedded into the cover data according to the matrix. In this paper, we propose a general framework for ME based on covering sequence (CS) of finite abelian group. By the proposed approach, the to-be-embedded message is regraded as an element of a finite abelian group, and it can be embedded into the cover data according to a CS of the group. It can be verified that many previous works, including the conventional ME (binary and ternary) and the sum and difference covering set based steganography, are special cases of the proposed general framework. The proposed CS-based ME formally extends these classical algorithms, and it provides a general way for designing efficient steganography. Some examples of CS-based ME and their performance evaluation are also given for a better illustration.
An efficient method of data hiding for digital colour images based on variant expansion and modulus function
2021, Journal of Engineering Science and Technology

View all citing articles on Scopus

View full text

Steganalysis of a PVD-based content adaptive image steganography

Abstract

Highlights

Introduction

Section snippets

Embedding procedure of Luo et al.'s method

A case study with the block size Bz=1

Experimental results

Conclusion

Acknowledgments

Signal Processing

Signal Processing

Pattern Recognition Letters

Pattern Recognition Letters

Steganography in Digital MediaPrinciples, Algorithms, and Applications

A survey on image steganography and steganalysis

Journal of Information Hiding and Multimedia Signal Processing

Image steganographic scheme based on pixel-value differencing and LSB replacement methods

IEE Proceedings Vision, Image & Signal Processing

Adaptive data hiding in edge areas of images with spatial LSB domain systems

IEEE Transactions on Information Forensics and Security