The ISC International Journal of Information Security

Abstract Least Significant Bit Matching (LSBM) is a simple steganography approach that has been detected under multiple attacks. Imperceptibility (i.e., maintenance of high perceptual image quality) and security are significant parameters in steganography. However, most conventional steganography techniques rely on single-objective optimization, which focuses on improving one parameter while often compromising others. This limitation underscores the need for approaches that balance conflicting objectives. To address this, the present study employs the Non-Dominated Sorting Genetic Algorithm II (NSGA-II) to optimize security and imperceptibility. This methodology includes a cover image division into blocks, each with two critical decisions: (1) seed determination for the pseudo-random number generator to simultaneously identify optimal pixels for data embedding and (2) selecting whether the pixel value should be increased or reduced upon a mismatch between the data bit and pixel LSB. Pixels with the highest data bit–LSB correspondence are optimal, and a pixel value change (increase or reduction) is to minimize block histogram variation. This multiobjective optimization is carried out using NSGA-II. It was comparatively revealed that the developed methodology remarkably improved image quality metrics and decreased detection accuracy at different embedding rates. At embedding rates of 0.3, 0.5, and 0.8 bpp, the Peak Signal-to-Noise Ratio (PSNR) was approximately 57.65, 55.55, and 52.75, respectively. This result represents a 1.5-2.5% improvement compared to conventional LSBM techniques. 

Abstract A steganography system must embed the message in an unseen and unrecognizable manner in the cover signal. Embedding information in transform coefficients, especially Discrete Wavelet Transform (DWT), is one of the most successful approaches in this field. The proposed method in this paper has two main steps. In the first step, the XOR logical function was used to embed two bits of data in the adjacent DWT coefficient pair. No change in the coefficients will occur if the XOR result of the two bits of low-value data of the two adjacent coefficients is identical to the two bits of secret data. Otherwise, one or both of the coefficient(s) will need a one-unit increase or decrease. In the second step, the genetic algorithm was used to select, between the two possible solutions, a new value for the adjacent coefficient pair that needs to be changed. Using the genetic algorithm, the selections were made such that the generated stego image experienced the least change relative to the cover image. The results of comparing this method with the existing methods in low- and high-level embedding showed that the proposed method was successful in producing stego images with high-quality criteria. In addition, the SPAM steganalysis method did not show high accuracy in its detection. One of the benefits of the proposed method is the need for a short key to embed and extract the secret message. This issue increases the security and feasibility of the proposed method.

Abstract Steganalysis is an interesting classification problem to discriminate the images, including hidden messages from the clean ones. There are many methods, including deep CNN networks, to extract fine features for this classification task. Also, some researches have been conducted to improve the final classifier. Some state-of-the-art methods use ensemble of networks by a voting strategy to achieve more stable performance. In this paper, a selection phase is proposed to filter improper networks before any voting. This filtering is done by a binary relevance multi-label classification approach. Xu-Net and ResT-Net, the most famous state-of-the-art Steganalysis ensemble models, are considered as the base networks for feature extraction. The Logistic Regression (LR) is chosen here as the last layer of the networks for classification. One large-margin Fisher's linear discriminant (FLD) classifier is trained for each one of the networks to measure its suitability in classifying the query image. The proposed method with different approaches is applied on the BOSSbase dataset and compared to traditional voting and some state-of-the-art related ensemble techniques. The results show significant accuracy improvement of the proposed method in comparison with others.

Abstract In network steganography methods based on packet length, the length of the packets is used as a carrier for exchanging secret messages. Existing methods in this area are vulnerable against detections due to abnormal network traffic behaviors. The main goal of this paper is to propose a method which has great resistance to network traffic detections. In the first proposed method, the sender embeds a bit of data in each pair that includes two non-identical packet lengths. In the current situation, if the first packet length of the pair is larger than the second one, it shows a ‘1’ bit and otherwise, it shows a ‘0’ bit. If the intended bit of the sender is in conflict with the current status, he/she will create the desired status by swapping the packet lengths. In this method, the paired packets can be selected freely, but in the second proposed method, the packets are divided into buckets and only packets within a single bucket can be paired together. In this case, the embedding method is similar to the previous one. The results show that the second method, despite having low embedding capacity, will be more secure in real traffic compared to the other methods. Since the packet lengths of UDP protocol are more random in comparison to TCP, the proposed methods have higher embedding capacity and they are more secure for UDP-based packets. However, these methods are only applicable to the protocols in which the packet length has not a constant value.

Abstract This paper investigates digital data hiding schemes. The concept of information hiding will be explained at first, and its traits, requirements, and applications will be described subsequently. In order to design a digital data hiding system, one should first become familiar with the concepts and criteria of information hiding. Having knowledge about the host signal, which may be audio, image, or video and the final receiver, which is Human Auditory System (HAS) or Human Visual System (HVS), is also beneficial. For the speech/audio case, HAS will be briefly reviewed to find out how to make the most of its weaknesses for embedding as much data as possible. The same discussion also holds for the image watermarking. Although several audio and image data hiding schemes have been proposed so far, they can be divided into a few categories. Hence, conventional schemes along with their recently published extensions are introduced. Besides, a general comparison is made among these methods leading researchers/designers to choose the appropriate schemes based on their applications. Regarding the old scenario of the prisoner-warden and the evil intention of the warden to eavesdrop and/or destroy the data that Alice sends to Bob, there are both intentional and unintentional attacks to digital information hiding systems, which have the same effect based on our definition. These attacks can also be considered for testing the performance or benchmarking, of the watermarking algorithm. They are also known as steganalysis methods which will be discussed at the end of the paper.

Abstract So far, various components of image characteristics have been used for steganalysis, including the histogram characteristic function, adjacent colors distribution, and sample pair analysis. However, some certain steganography methods have been proposed that can thwart some analysis approaches through managing the embedding patterns. In this regard, the present paper is intended to introduce a new analytical method for detecting stego images, which is robust against some of the embedding patterns designed specifically to foil steganalysis attempts. The proposed approach is based on the analysis of the eigenvalues of the cover correlation matrix used for the purpose of the study. Image cloud partitioning, vertical correlation function computation, constellation of the correlated data, and eigenvalues examination are the major challenging stages of this analysis method. The proposed method uses the LSB plane of images in spatial domain, extendable to transform domain, to detect low embedding rates-a major concern in the area of the LSB steganography. The simulation results based on deviation detection and rate estimation methods indicated that the proposed approach outperforms some well-known LSB steganalysis methods, specifically at low embedding rates.

Abstract Perturbed Quantization (PQ) steganography scheme is almost undetectable with the current steganalysis methods. We present a new steganalysis method for detection of this data hiding algorithm. We show that the PQ method distorts the dependencies of DCT coefficient values; especially changes much lower than significant bit planes. For steganalysis of PQ, we propose features extraction from the empirical matrix. The proposed features can be exploited within an empirical matrix of DCT coefficients which some most significant bit planes were deleted. We obtain four empirical matrices and fuse resulted features from these matrices which have been employed for steganalysis. This technique can detect PQ embedding on stego images with 77 percent detection accuracy on mixed embedding rates between 0.05 _ 0.4 bits per non-zero DCT AC coefficients (BPNZC). Comparing the results, we also show that the detection rates are effectively comparable with respect to current steganalysis techniques for PQ steganography.

Abstract In this paper a steganalysis method is proposed for pixel value differencing method. This steganographic method, which has been immune against conventional attacks, performs the embedding in the difference of the values of pixel pairs. Therefore, the histogram of the differences of an embedded image is di_erent as compared with a cover image. A number of characteristics are identified in the difference histogram that show meaningful alterations when an image is embedded. Five distinct multilayer perceptrons neural networks are trained to detect different levels of embedding. Every image is fed in to all networks and a voting system categorizes the image as stego or cover. The implementation results indicate an 88.6% success in correct categorization of the test images.