Subtractive Pixel Adjacency Matrix Research Articles

Towards improving the performance of blind image steganalyzer using third-order SPAM features and ensemble classifier

The success rate for blind or universal steganalysis lies in the ability to extract the statistical footprints of image features. Further, the choice of machine learning (ML) algorithm is crucial to distinguish the stego image more precisely from the untouched clean images. Literature suggests that most steganalysis approaches report less favorable detection accuracy despite considering many features. This study presents a three-step process to accurately identify the clean and stego images to solve this issue. We used the curvelet denoising as an initial phase during the first step to suppress the natural noise residuals (NRs) by producing the stego NRs. Secondly, it extracts the Third-order Markov-chain sample transition probability matrices as features. Finally, the oblique decision tree ensemble using a multisurface proximal support vector machine (SVM) classifier has been utilized to achieve greater detection accuracy than the state-of-the-art classifiers. The experiments are performed on an extensive database comprising clean and stego images generated from nine embedding schemes with varying payloads. The experimental results suggest that an accuracy of 93.12 has been achieved using the proposed Third order subtractive pixel adjacency matrix (SPAM) features with an ensemble classifier.

One-Class Double Compression Detection of Advanced Videos Based on Simple Gaussian Distribution Model

Passive video forensics has become an active topic in recent years. Generally, a pristine video obtained from surveillance cameras or other video recording devices is single compressed. At the same time, double lossy compression will certainly be introduced in tampered videos since it needs to go through recompression to perform tampering on a commonly used compressed video. In the video’s double compression, some traces are left due to the intrinsic effects of recompression. Traditional supervised learning is inefficient because it requires two classes (the pristine and the manipulated) that occur in the video to be exhaustively assigned labels. Actually, compared with pristine videos, manipulated videos with labels are more difficult to obtain. To address this problem, in this paper, one-class classification, which is often used for anomaly detection and only needs the target class, is introduced. We first treat all decompressed video frames as still images and extract subtractive pixel adjacency matrix (SPAM) steganalysis features to detect traces left in the double compression process. Then we adopt a Gaussian density-based one-class classifier since SPAM features extracted from pristine video frames approximately subject to the Gaussian distribution. Furthermore, we improve the robustness of the classifier by using ensemble strategy. Experimental results indicate that our proposed method exceeds other more complex one-class classification methods, and outperforms fully-supervised learning methods only by feeding features from single compressed video frames to our one-class classifier.

Multi-directional block based PVD and modulus function image steganography to avoid FOBP and IEP

Since the inception of pixel value differencing (PVD) image steganography, it has drawn considerable interest among the researchers of this field. However, most of the PVD based techniques suffer from either falling-off boundary problem (FOBP) or incorrect extraction problem (IEP). Therefore, to address these two issues, this paper proposes a multi-directional pixel value differencing and modulus function (MDPVDMF) based technique. During the embedding process, the original image (OI) is partitioned into 2 × 2 size pixel blocks. Then, data embedding is performed by exploiting the horizontal, vertical, and diagonal directions for each block. For a 2 × 2 pixel block, two difference values can be obtained in any of the three directions. Next, using the difference values and the remainders of the pixel pairs, the secret bits are embedded. The experiment has been conducted to compute the performance of the proposed technique with regards to the image quality metrics like peak signal-to-noise ratio (PSNR), embedding capacity (EC), and FOBP. Results show that PSNR is optimal for vertical pairs with 39.17 dB whereas the EC is optimal for the diagonal pairs with 3.10 bits per pixel (BPP). Further, the proposed technique has shown exceptional attack resistance ability to regular & singular (RS) attack, salt & pepper (S&P) noise, pixel difference histogram (PDH) analysis, and subtractive pixel adjacency matrix (SPAM) steganalysis.

Improving Image Steganalyser Performance using Second Order SPAM Features Extracting through Contourlet Transform

The major challenge posed by feature based blind steganalysers is the scheming of useful image features, which offers true existence of the stego noise rather than the natural noise in the images. Despite hundreds of features being applied in the real time implementation, only low detection accuracy could be achieved. Hence, this paper proposes a new model for detecting the stego image coupled with an examination of the task by applying a two-step process. (a) Extraction of the second order SPAM (Subtractive Pixel Adjacency Matrix) as features and the second order SPAM features of coefficients and co-occurrence matrices of sub band images from the contourlet transform. (b) Implementation of the system, based on an efficient classifier, Support Vector Machine which is capable of providing the higher detection rate than the existing classifers. Full- fledged experimentation with huge database of clean and steganogram images produced from seven steganographic schemes with varying embedding rates, and using five steganalysers were carried out in this study. The study shows that the proposed paradigm enhances the detection accuracy rate substantially and validates its efficiency with its better performance even at low embedding rates.

A New Steganography Method Using Integer Wavelet Transform and Least Significant Bit Substitution

Abstract Steganography is a data hiding technique, which is used for securing data. Both spatial and transform domains are used to implement a steganography method. In this paper, a novel transform domain method is proposed to provide a better data hiding method. The method uses a multi-resolution transform function, integer wavelet transform (IWT) that decomposes an image into four subbands: low-low, low-high, high-low and high-high subband. The proposed method utilizes only the three subbands keeping the low-low subband untouched which helps to improve the quality of the stego image. The method applies a coefficient value differencing approach to determine the number of secret bits to be embedded in the coefficients. The method shows a good performance in terms of embedding capacity, imperceptibility and robustness. A number of metrics are computed to show the quality of the stego image. It can also withstand RS steganalysis, Chi-squared test and Subtractive Pixel Adjacency Matrix steganalysis successfully. The deformation of the histogram and Pixel Difference Histogram for different embedding percentages are also demonstrated, which show a significant similarity with the original cover image. The proposed method shows an achievement of 2.3bpp embedding capacity with a good quality of stego image.

Integration of statistical detector and Gaussian noise injection detector for adversarial example detection in deep neural networks

Deep Neural Networks (DNN) has achieved a great success in many tasks in recent years. However, researchers found that DNN is vulnerable to adversarial examples that are maliciously perturbed inputs. The elaborately designed adversarial perturbations can easily confuse the model whereas have no impacts on human perception. To counter adversarial examples, we propose an integrated detection framework for detecting adversarial examples, which involves statistical detector and Gaussian noise injection detector. The statistical detector extracts Subtractive Pixel Adjacency Matrix (SPAM) and uses the second order Markov transition probability matrix to model SPAM so as to highlight the statistical anomaly hidden in an adversarial input. Then an ensemble classifier using SPAM based feature is applied to detect the adversarial input containing large perturbation. The Gaussian noise injection detector first injects an additive Gaussian noise into the input, and then feeds both the original input and its Gaussian noise injected counterpart into a targeted network. By comparing the two outputs difference, the detector is applied to detect adversarial input containing small perturbation: if the difference exceeds a threshold, the input is adversarial; otherwise legitimate. The two detectors are adaptive to different characteristics of adversarial perturbation so that the proposed detection framework is capable of detecting multiple types of adversarial examples. In our work, we test six categories of adversarial examples produced by Fast Gradient Sign Method (FGSM, untargeted), Randomized Fast Gradient Sign Method (R-FGSM, untargeted), Basic Iterative Method (BIM, untargeted), DeepFool (untargeted), Carlini&Wagner Method (CW_UT, untargeted) and CW_T(targeted). Comprehensive empirical results show that the proposed detection framework has achieved a promising performance on ImageNet database.

A study on Subtractive Pixel Adjacency Matrix features

Subtractive Pixel Adjacency Matrix (SPAM) features perform well in detecting spatial-domain steganographic algorithm. Further, some methods of SPAM features can be applied to rich models and steganalysis based on deep learning. Therefore, this paper presents a study on SPAM features and it is divided into two parts: in the first part, impact of spatial-domain steganographic on difference between adjacent pixels is first analyzed. Then, three SPAM features are proposed with the same range of differences and different orders of Markov chain. Following that, the influences of order of Markov chain and range of differences on SPAM features are analyzed, and we find that detection accuracy of SPAM features increases with the range of differences increasing; in the second part, SPAM feature is first divided into several modules according to the conclusion. Then, taking detection accuracy of support vector machine (SVM) classifier and mutual information as metrics and module as a unit, a Novel Feature Selection (NFS) algorithm and an Improved Feature Selection algorithm are proposed. Experimental results show that the NFS algorithm can achieve higher detection accuracy than several existing algorithms.

An optimised approach to detect the identity of hidden information in grey scale and colour images

Feature-based steganalysis is an emerging trend in the domain of Information Forensics, aims to discover the identity of secret information present in the covert communication by analysing the statistical features of cover/stego image. Due to massive volumes of auditing data as well as complex and dynamic behaviours of steganogram features, optimising those features is an important open problem. This paper focused on optimising the number of features using the proposed quick artificial bee colony (qABC) algorithm. Here we tested for three steganalysers, namely subtractive pixel adjacency matrix (SPAM), phase aware projection model (PHARM) and colour filter array (CFA) for the break our steganographic system (BOSS) 1.01 datasets. The significant improvement in the convergence nature of qABC quickly improves the solution and fine tune the search than their real counterparts. The results reveal that qABC method with support vector machine (SVM) classifier outperforms the non-optimised version concerning classification accuracy and reduced number of feature sets.

PFA-based feature selection for image steganalysis

This paper presents a universal image steganalysis method based on feature selection by principal feature analysis (PFA). The goal of this paper is to increase the performance of existing image steganalysis approaches using PFA-based feature selection method and reduce the high dimensionality of the features used in state-of-the-art steganalysis methods. Principal component analysis (PCA) is widely used in pattern recognition applications. However, PCA has disadvantage that, all the generated features are transformed features. While, PFA selects the subset of preliminary features which contains necessary information. PFA is applied on spatial domain subtractive pixel adjacency matrix features and in case of transform domain, CHEN features (intrablock and interblock Markov-based features) and CC-PEV features (PEV features enhanced by Cartesian calibration). The experimental results show that PFA is effective and efficient in eliminating redundant features. Experimental results prove that the use of PFA method in steganalysis is superior in terms of dimensionality reduction of features and increases the classification performance.

PFA-based feature selection for image steganalysis

This paper presents a universal image steganalysis method based on feature selection by principal feature analysis (PFA). The goal of this paper is to increase the performance of existing image steganalysis approaches using PFA-based feature selection method and reduce the high dimensionality of the features used in state-of-the-art steganalysis methods. Principal component analysis (PCA) is widely used in pattern recognition applications. However, PCA has disadvantage that, all the generated features are transformed features. While, PFA selects the subset of preliminary features which contains necessary information. PFA is applied on spatial domain subtractive pixel adjacency matrix features and in case of transform domain, CHEN features (intrablock and interblock Markov-based features) and CC-PEV features (PEV features enhanced by Cartesian calibration). The experimental results show that PFA is effective and efficient in eliminating redundant features. Experimental results prove that the use of PFA method in steganalysis is superior in terms of dimensionality reduction of features and increases the classification performance.

A proactive secret image sharing scheme with resistance to machine learning based steganalysis

In secret image sharing (SIS) schemes, a secret image is shared among a set of n images called stego-images. Each stego-image is preserved by a participant. In the recovery stage, at least k out of n stego-images are required to obtain the secret image, while k − 1 cannot reveal the secret in the sense of perfect secrecy. Hence, SIS guarantees long-term security. However, as the longer the stego-images remain stored, the higher is the probability of being vulnerable against steganalysis. To resolve this issue, this paper proposes the use of proactive secret sharing in an SIS scheme (P-SIS). P-SIS allows the stego-images to be renewed frequently while these are stored, without changing both cover and secret images. However, direct implementation of a proactive SIS requires more embedding rate (ER), causing high steganalysis accuracy detection and loss of quality in the stego-images. Our proposal addresses this issue and presents the combination of a (k, L, n)-threshold ramp secret sharing scheme and least significant bit matching (LSBM) steganography to reduce the steganalysis accuracy detection. The results of the evaluation show effectiveness of the proposal in terms of good quality of the stego-images, accurate recovery of the secret, and reduce the ER. Note that, despite the extensive research of SIS presented until now, only a few previous work is found on steganalysis in SIS. Not only constructing P-SIS scheme, but we also experimented the tolerance of the proposed P-SIS scheme against stganalysis in this paper. As a result, it is shown that the proposed scheme can withstand steganalysis based on machine learning (i.e., based on subtractive pixel adjacency matrix, SPAM).

An optimised approach to detect the identity of hidden information in gray scale and colour images

Feature-based steganalysis is an emerging trend in the domain of Information Forensics, aims to discover the identity of secret information present in the covert communication by analysing the statistical features of cover/stego image. Due to massive volumes of auditing data as well as complex and dynamic behaviours of steganogram features, optimising those features is an important open problem. This paper focused on optimising the number of features using the proposed quick artificial bee colony (qABC) algorithm. Here we tested for three steganalysers, namely subtractive pixel adjacency matrix (SPAM), phase aware projection model (PHARM) and colour filter array (CFA) for the break our steganographic system (BOSS) 1.01 datasets. The significant improvement in the convergence nature of qABC quickly improves the solution and fine tune the search than their real counterparts. The results reveal that qABC method with support vector machine (SVM) classifier outperforms the non-optimised version concerning classification accuracy and reduced number of feature sets.

An improved dynamic discrete firefly algorithm for blind image steganalysis

This paper presents an improved firefly algorithm (DyFA) for feature selection that improves the convergence rate and reduces computational complexity through dynamic adaptation in blind image steganalysis. The alpha and gamma parameters of the Firefly algorithm are made to vary dynamically with each generation for faster convergence. If firefly algorithm’s performance does not improve for certain numbers of iterations then the particles with the worst fitness function values are replaced with new particles in the search space and particle dimensions are reduced by eliminating redundant features. This approach is effective in reducing computational complexity and improving detection capability of the classifier. To further reduce the computational complexity a hybrid DyFA is designed by ensemble of a filter approach (t test + regression) and wrapper approach (DyFA) incrementally. In this study, support vector machine classifier with radial basis function kernel and ten fold cross validation is used to evaluate the effectiveness of the proposed Firefly algorithm. DyFA is compared with well-known wrapper feature selection algorithms. Experimental results are performed on datasets constructed from four steganography algorithms nsF5, Perturbed Quantization, Outguess and Steghide with subtractive pixel adjacency matrix (SPAM) feature vector from spatial domain and Cartesian Calibrated features extracted by Pevnýfeature vector from transform domain. Experimental results demonstrate that DyFA reduces computation time and improves classification accuracy as compared to other feature selection algorithms. Hybrid DyFA shows an improvement in classification accuracy and in eliminating redundant features in more than 85 % of cases with respect to hybrid GLBPSO.

디지털 영상 픽셀값의 경사도를 이용한 Downscaling Forgery 검출

스마트 기기와 소형 디스플레이에 사용되는 디지털 영상은 다운스케일링 (Downscaling)된 영상이 사용된다. 본 논문에서는 영상 픽셀값의 경사도에 따른 특징벡터를 이용한 다운스케일링 포저리 (Forgery) 영상 검출 알고리즘을 제안한다. 제안된 알고리즘에서, 원영상의 픽셀값 경사도로부터 자기회귀 (AR: Autoregressive) 계수를 계산한다. 이는 다운스케일링 포저리 영상검출기의 SVM (Support Vector Machine) 분류를 위한 학습에 사용된다. 제안된 다운스케일링 검출 알고리즘은 동일 10-Dim. 특징벡터의 MFR (Median Filter Residual) 스킴과 686-Dim.의 SPAM (Subtractive Pixel Adjacency Matrix) 스킴과 비교하여 다운스케일링 90% 영상 포저리에서 성능이 우수하며, 평균필터링 (3×3) 영상과 미디언필터링 (3×3) 영상에서 높은 검출율을 보여 주었다. 특히, 평균필터링과 미디언필터링 영상에서는 성능평가 전체 항목에서 민감도 (Sensitivity; TP: True Positive rate)와 1-특이도 (1-Specificity; FP: False Positive rate)의 AUC (Area Under Curve)가 모두 1에 수렴하여 ‘Excellent (A)’ 등급임을 확인하였다.

Multi-Bit Data Hiding Scheme for Compressing Secret Messages

The goal of data hiding techniques usually considers two issues, embedding capacity and image quality. Consequently, in order to achieve high embedding capacity and good image quality, a data hiding scheme combining run-length encoding (RLE) with multi-bit embedding is proposed in this paper. This work has three major contributions. First, the embedding capacity is increased 62% because the secret message is compressed before embedding into the cover image. Secondly, the proposed scheme keeps the multi-bit generalized exploiting modification direction (MGEMD) characteristics, which are effective to reduce modified pixels in the cover image and to maintain good stego image quality. Finally, the proposed scheme can prevent modern steganalysis methods, such as RS steganalysis and SPAM (subtractive pixel adjacency matrix), and is compared to MiPOD (minimizing the power of the optimal detector) scheme. From our simulation results and security discussions, we have the following results: First, there are no perceivable differences between the cover images and stego images from human inspection. For example, the average PSNR of stego images is about 44.61 dB when the secret message (80,000 bits) is embedded for test cover images (such as airplane, baboon, Lena) of size 512×512. Secondly, Appl. Sci. 2015, 5 1034 on average, 222,087 pixels were not modified after embedding for the cover image. That is to say, 12% less pixels are modified as compared to the MGEMD method. From the performance discussions, the proposed scheme achieves high embedding capacity and good image quality, but also maintains stego image security.

Steganalysis of images based on spatial domain and two-dimensional JPEG array

The aim of digital image steganalysis is to detect hidden information (which can be a message or an image) in a steganographic image. An ideal steganography method encrypts the information in the image such that it cannot be easily detected. Currently, a wide variety of different steganography techniques are being used; therefore, more advanced steganalysis methods are needed that can detect the steganographic images coded by different techniques. A typical steganalysis technique consists of two parts: (1) feature extraction and (2) classification. In this paper, a new steganalysis technique based on the Markov chain process is proposed. In the proposed technique, after extraction of the new features, a non-linear classifier named support vector machine is applied to classify clean and encrypted images. Analysis of variance is used to reduce the dimensions of the proposed features. The performance of the proposed technique is compared against subtractive DCT coefficient adjacency matrix (SDAM) and subtractive pixel adjacency matrix (SPAM) methods using an image database prepared by three strong steganography techniques called yet another steganographic scheme, model based, and perturbed quantization. The obtained results show that the proposed method provides better performance than SDAM and SPAM methods.

Selection of image features for steganalysis based on the Fisher criterion

A steganalytic feature selection method based on the Fisher criterion used in pattern recognition is proposed in this paper in order to reduce effectively the high dimensionality of the statistical features used in state-of-the-art steganalysis. First, the separability of each single-dimension feature in the feature space is evaluated using the Fisher criterion, and these features are reordered in descending order of separability. Then, starting from the first dimension of the reordered features, as the dimension increases, the separability of each feature component is analyzed using the Fisher criterion combined with the Euclidean distance. Finally, the feature components with the best separability are selected as the final steganalytic features. Experimental results based on the selection of SPAM (Subtractive Pixel Adjacency Matrix) features in spatial-domain steganalysis and CC-PEV (Cartesian Calibrated feature extracted by PEVný) features in DCT-domain steganalysis show that the proposed method can not only reduce the dimensionality of the features efficiently while maintaining the accuracy of the steganalysis, but also greatly improve the detection efficiency.

Steganalysis by subtractive pixel adjacency matrix and dimensionality reduction

Subtractive pixel adjacency matrix (SPAM) features, introduced by Pevn’y et al. as a type of Markov chain features, are widely used for blind steganalysis in the spatial domain. In this paper, we present three improvements to SPAM as follows: 1) new features based on parallel subtractive pixels are added to the SPAM features, which only refer to collinear subtractive pixels; 2) features are extracted not only from the spatial image, but also from its grayscale-inverted image, making the feature matrices symmetrical and reducing their dimensionality by about half; and 3) a new kind of adjacency matrix is used, thereby reducing about 3/4 of the dimensionality of the features. Experimental results show that these methods for dimensionality reduction are very effective and that the proposed features outperform SPAM.

Steganalysis by Subtractive Pixel Adjacency Matrix

This paper presents a method for detection of steganographic methods that embed in the spatial domain by adding a low-amplitude independent stego signal, an example of which is least significant bit (LSB) matching. First, arguments are provided for modeling the differences between adjacent pixels using first-order and second-order Markov chains. Subsets of sample transition probability matrices are then used as features for a steganalyzer implemented by support vector machines. The major part of experiments, performed on four diverse image databases, focuses on evaluation of detection of LSB matching. The comparison to prior art reveals that the presented feature set offers superior accuracy in detecting LSB matching. Even though the feature set was developed specifically for spatial domain steganalysis, by constructing steganalyzers for ten algorithms for JPEG images, it is demonstrated that the features detect steganography in the transform domain as well.