This research article provides an up-to-date review of spatial-domain steganography. Maintaining the communication as secure as possible when transmitting secret data through any available communication channels is the target of steganography. Currently, image steganography is the most developed field, with several techniques are provided for different image formats. Therefore, the general image steganography including the fundamental concepts, the terminology, and the applications are highlighted in this paper. Further, the paper depicts the essential characteristics between information hiding and cryptography systems. In addition, recent well-known techniques in the spatial-domain steganography, such as LSB and pixel value differencing, are discussed in detail and several comparisons are provided to show the merits and the demerits of the discussed techniques. Furthermore, to aid the steganography researchers in developing efficient spatial-domain embedding techniques, the future research of the spatial-domain steganography is discussed and a set of recommendations are suggested.
Steganography is a word that is derived from the Greek word, namely, “Stegos”, which refers to the word “cover”, and the word “Grafia” refers to the word “writing”, which is defined as “covered writing” [
Information hiding contains two sub domains, which are watermarking and steganography [
Information Hiding | Cryptography (Encryption) | ||
---|---|---|---|
Watermarking | Steganography | ||
Main objective | Protect media copyrights | Conceal existence of secret data and communication | Content protection |
Robustness | Against removing tampering security data | Against detecting the existence of secret data | Against breaking ciphers |
Secret information | Watermark | Payload | Plain text of file |
Security of communication | Depends on how confidential the embedding method is. | Depends on how confidential the embedding method is. | Depends on how confidential the key is. |
Loss of security | When loss of integrity | When detecting the existence of security data | When decrypting the cipher |
Result | Watermarked media | Stego-media | Cipher |
Need for key | Optional | Depends on used application | Compulsory |
Type of attacks | Image processing | Steganalysis | Cryptanalysis |
Main challenges | Robustness | Imperceptibility, embedding payload, and robustness | complexity of encryption and key management |
New review papers in the domain of image steganography are always needed to discuss the strengths and limitations of recent proposed techniques. Therefore, this review paper aims to present an up-to-date knowledge for spatial-domain image steganography. The rest of the paper is structured as follows: Section 2 presents an overview of the steganography including history, fundamental concepts, terminology, and applications. Section 3 reviews the current spatial-domain image steganography, where recent well-known techniques are analyzed and discussed in detail. The future research and recommendations, which could forward the researchers to develop efficient spatial-domain embedding schemes, are presented in Section 4. Finally, Section 5 concludes the paper.
The notion of making any information hidden from other unauthorized users has been studied since the 440 BC and applied into various layouts through the past few years ago [
Most of the current steganographic systems make use of different multimedia objects such as images, videos, audio files, and so on, as a covering media since digital images are frequently transmitted by users through emails and many different communications [
Modern steganographic systems apply the inventions of the networking and computers that appeared in the 20th century. There exist four major improved aspects regarding the digital steganography. These comprise network steganography, file system steganography, linguistic steganography and digital media steganography [
The term cover image indicates to the image that is used for the purpose of transporting the embedded bits [
In
where the size of the secret message |SM|, hidden in
An image is a frequently applied file format in steganography since a secret message is entirely embedded within a cover image. Image steganography is categorized into transform and spatial domains. In the spatial domain, a secret message is efficiently embedded within a pixel value in a direct manner. In the transform domain, different approaches perform the embedding procedure based on initially transforming an image from a spatial domain along to a frequency domain based on the use of any available transforms. In the next step, the embedding procedure is applied according to appropriate transformed coefficients.
The measures of an image quality are applied in order to assess the quality of the stego-image that is acquired following the embedding procedure. Various approached aim to attack the steganographic algorithm. Several techniques related to steganography are existing where such techniques involve StegHide, Hide4PGP, S-tools, OutGuess, Stegnos, Ezstego, Hide and Seek, F5, Mp3Stego, and so on. Many different methods of steganalysis comprise the efficient selection of any available secret message and detecting its estimated length or a way of retrieving it. Many different stego attacks involve the filter attack, image resizing attack, J. Fridrich’s RS steganalysis, JPEG compression attack, Chi Square attack, Jeremiah J. Harmsena’s Histogram attack, image tampering attack, AWGN attack, and so on. The algorithm that is applied to embed the secret data must endure the whole attacking types by preventing an eavesdropper from obtaining the hidden message [
Steganography is applied into many different suitable applications [
The indispensable characteristic of steganography is based on maintaining the communication as secure as possible when transmitting the stego-image through any available communication or networking channels [
The spatial domain can utilize the cover image pixels for concealing the secret information, such as the replacement of secret bits within a pixel value [
Moreover, the adaptive embedding represents a statistical or model-based approach that manages different methods related to information hiding. In fact, this approach is interwoven to the transform and spatial domains. This embedding method’s type is based on considering the statistical characteristics of an image prior to applying the embedding procedure. This set of the statistical characteristics dictates where the modifications take place in the cover image [
The spatial-domain embedding techniques are more common in comparison with the transform domain due to its simplicity in the embedding and extraction procedures, but with less strength [
Characteristics | Spatial Domain | Transform Domain | Adaptive Embedding |
---|---|---|---|
Embedding Capacity (Payload) | High | Low | Vary, method dependent |
Embedding place | Direct manipulating of pixel values | Transform coefficients | Method dependent |
Cover format dependency | Format Dependent | Format independent | Method dependent |
Complexity | Low | High | Method dependent |
Robustness against noise, compression, cropping, etc. | Not robust | Less prone | Method dependent |
Visual Quality (Imperceptibility) | High | Low | Low |
Geometric attacks | Not robust (vulnerable) | Less prone (resistant) | Less prone (resistant) |
Statistical detection attacks (e.g., Histogram, RS-attack) | Easy to detect | Hard to detect | Hard to detect |
Well-known techniques | LSB, PVD, MBNS steganography | DCT based, DWT based, CWT based steganography | Region based, HVS, Machine Learning and AI based steganography |
The simplest method of conducting the process of data embedding through digital images is based on updating the values of cover pixels within the spatial domain [
Approach | Reference | Method Name | Merits | Challenges | Payload (bpp) | Visual Quality (PSNR) | Resistance against Steganalysis |
---|---|---|---|---|---|---|---|
LSB | Sarreshtedari et al. [ |
±1 LSB | High imperceptibility & simple implementation. | Lower payload & Key dependent | 1 bpp (gray-image) | ~53 dB | HCF-COM |
Qazanfari et al. [ |
GLSB++ | Secure against Histogram analysis & improved visual quality. | Not robust & key dependent | 0.8 bpp (gray-image) | >50 dB | Chi-Square & Histogram | |
Nguyen et al. [ |
MPBDH | Adaptive embedding & reduce visual attacks. | Not robust against compression & cropping & key dependent | ~1.5 bpp (gray-image) | ~46 dB | SPAM at low embedding | |
Muhammad et al. [ |
MLEA | Keeps balance between imperceptibility and security, & applies multi-level of encryption for secret data. | Lower payload | ~1 bpp | >45 dB | Salt & pepper noise, & Histogram | |
Rajendran and Doraipandian [ |
logistic map-LSB | High visual quality & & simple implementation. | Lower payload | 2 bpp | >44.5 dB | Histogram | |
Vyas and Dudul [ |
OO-LSB | Encrypts the secret data before embedding starts & embeds in skin areas. | Uses multiple covers | ~40KB | >47 dB | N/A | |
PVD | Balasubra manian et al. [ |
Octonary-PVD | Adaptive embedding & resistance against various statistical steganalysis. | Modern steganalysis evaluation is missing | ~ 3.6 bpp (gray-image) | ~40 dB | PVD analysis & RS analysis |
Shen et al. [ |
MF-PVD | Resolves the PVD underflow/overflow problem & simple implementation. | Limited payload & modern steganalysis evaluation is missing | ~1 bpp (color-image) | ~36 dB | Pixel Difference |
|
Swain [ |
Ad-PVD | Adaptive embedding. | Lower payload | ~1.74 bpp |
~46.7 dB | Pixel Difference |
|
GrajedaMarín et al. [ |
PVD-TPVD | Resolves the PVD underflow/overflow & embedding done by full utilization of pixels. | Security evaluation by steganalysis is missing | ~2.14 bpp (gray-image) | ~38.3 dB | N/A | |
Swain [ |
Ad-PVD | Adaptive embedding & High visual quality | Complex algorithms for embedding & extracting | ~3 bits per byte | ~43 dB | Pixel Difference |
|
EMD | Kuo et al. [ |
GEMD | Uses dynamic modulus table to resolve the extraction function fixed weighting problem. | Modification of all pixels to embed the secret data | 1.5 bpp | ~50.2 dB | N/A |
Kuo, Wang, et al. [ |
MSD | Maintains the bpp with increasing of n pixels & reduces the pixel modification ratio (only n/2 of pixels modification). | Limited payload | Only 1 bpp | >52 dB | RS analysis | |
Kuo et al. [ |
MBEF | Adaptive embedding & resolves the PVD underflow/overflow. | Low visual quality when high payload is embedded | Bet. 1.25 & 4.5 bpp | Bet. 51 to 30 dB based on the payload | Bit plane & RS analysis | |
MBNS | Geetha et al. [ |
VRNS | Good visual quality. | Not robust against compression, filtering & cropping, limited payload | Only 1 bpp | ~41 dB | RS analysis |
Chen et al. [ |
GMB | Adaptive technique & increases security by coefficient mapping | SPAM analysis detection when payload is > 1 bpp | Bet. 1.46 to 3.8 bpp | Bet. 50 to 35 dB | SPAM analysis, Histogram, RS analysis | |
Nyeem [ |
Bit Plan Sclicing | high payload with high imperceptibility | Not robust against attacks | Bet. ~2.5 to ~7.3 bpp | ~57 dB at 2.5 bpp | Histogram | |
GLM | Muhammad et al. [ |
GLM-MLE | High imperceptibility & robust against salt & pepper. | Limited payload | 8 KB | ~57 dB | N/A |
Palette | Imaizumi et al. [ |
k-bit palette | Higher payload with higher visual quality | Location map is required to extract the embedded bits | Bet. 1 to 3 bpp | ~40 dB at 3 bpp | N/A |
Prediction | Jafar et al. [ |
MPE | Improved the prediction accuracy by using multiple predictors. | Limited payload & security evaluation by steganalysis is missing | Only 90574 bits | ~46 dB | N/A |
Benhfid et. al. [ |
MLBS | Good imperceptibility level | Limited payload | 1.8 bpp | ~40 dB | Chi-Square | |
Deep learning | Baluja [ |
CNN | Higher payload | Takes more time to embed the secret image and requires much more memory | 1:1 ratio | N/A | N/A |
Zhu et al. [ |
GAN | High extracting accuracy | Limited payload and requires excessive memory | 0.203 bpp | <40 dB for Combined model | ATS analysis | |
Shang et al. [ |
GANste | Better security | Limited payload | 0.4 bpp | <30 dB | FGSM & Onepixelattack |
The LSB technique is considered an extremely simple technique in its performance, and therefore, it represents one of those common spatial image steganographic techniques [
To achieve effectiveness of this technique, several developed LSB based image steganography versions are taken into account. The most significant versions apply different LSB matching algorithms [
The major benefit of the LSB steganography related to its ease of the embedding and extraction procedures. Nonetheless, LSB techniques are vulnerable to different statistical attacks, with some manipulations within the stego-image. As the LSB steganography represents the way of modifying the cover’s pixel values, its performance of extracting of the embedded data relies on some factors such as the compression quantization, noise effect, and intruder attacks.
Wu et al. presented a novel embedding aspect that relies on the occurring difference among pixel values [
Many enhanced PVD based image steganography versions were researched in order to improve the efficiency of PVD. The most significant versions apply the Adaptive PVD block technique by using different pseudo-random number techniques for determining the blocks [90] and tackling the fall-off boundary problem in the PVD technique [
The Exploiting modification direction (EMD) method is a common method that keeps the increased fidelity pertaining to the stego-images protected [
Kuo et al. [
A further spatial-domain embedding technique that relies on the Multiple Base Notational Systems (MBNSs) is proposed in order to transform the secret information through to the notational scheme prior to the embedding procedure [
In several techniques related to the MBNSs, secret information is changed into symbols and re-expressed according to the used MBNS (e.g., octal, decimal and binary systems) [
In [
Potdar et al. [
The Quantization Index Modulation (QIM) technique [
For enhancements based on the embedding payload and the reduction of distortion, several enhanced quantization-based steganography versions are researched. One of these versions utilizes the elastic indicators and adjacent correlation. Through this approach, the indexes are encoded based on the difference values that are derived from the neighbouring indexes and the elastic sub codebooks are applied to enhance the compression rate [
The Palette based steganography is proposed in [
Imaizumi et al. [
The prediction based embedding technique has currently attracted many researchers [
During the prediction step, a predictor is employed to estimate the pixel values of the host image. After that, the entropy coder is used to compress the prediction EV. The Median Edge Detector (MED) technique and the Gradient Adjusted Prediction (GAP) techniques represent new predictors that are applied in several prediction-based image coding techniques. Many different reversible prediction-based embedding techniques are enhanced and highlighted in the literature. Every technique attempts at enhancing many available proposed existing techniques.
Hong et al. [
Recently, Benhfid et al. [
In recent years, the introducing of deep learning in steganography has shown a great improvement in the effectiveness of steganography methods. Deep learning steganography is learned from machine learning. Several deep learning steganography methods [
Zhu et al. [
Recently, Shang et al. [
Many spatial-domain steganography techniques can achieve high payload, but they are susceptible to extremely few updates, which are likely to be encountered based on different image processing tasks (e.g., scaling, rotation, cropping, and so on). Moreover, these techniques recompense the image’s statistical features indicating a weak robustness towards image filters and lossy compression. As a summary,
Technique | Merits | Demerits |
---|---|---|
LSB | Acceptable payload & simple implementation | Not robust against statistical attacks and noise |
PVD | High payload with acceptable imperceptibility | Not robust against statistical attacks |
EMD | Better imperceptibility compared with LSB & PVD | Low payload |
MBNS | High payload & more robust against steganalysis process | Not robust against geometrical attacks |
GLM | High payload & low computational complexity | Not robust against attacks |
QIM | High payload | Prone to steganalysis & geometrical attacks |
Palette based | High payload & less distortion compared to other spatial-domain techniques | Not secure & it needs covers of specific lossless compression format |
Prediction based | Not prone to steganalysis attacks | Limited payload |
Deep learning based | Better imperceptibility and security | Limited payload |
The major challenges incurred in the spatial-domain image steganography comprise having high embedding payload and security, and having a lowest detectability [ The emphasis over the adaptive steganography: An adaptive approach represents the basic notion for obtaining an optimized method. Adaptiveness does not only develop the embedding efficiency but can as well protect the attempts of steganalysis with appropriate and efficient counter measures. The majority of the prediction and deep learning techniques are considered to represent the most effective selection for obtaining an adaptive nature to the system. This allows providing further improvements for all image steganography concepts starting from the imperceptibility along towards the embedding payload in comparison with different traditional embedding techniques. Statistics aware modelling: Due to the further improvements in steganalysis techniques, forming the most secure steganography method is getting more crucial. In order to form this method, the embedding secret data is added to particular regions instead of the whole image. These regions are called the Region of Interest (ROI). These regions must be determined based on applying the embedding procedure within the image’s portions that yield to obtain the lowest distortion. Consequently, it can be inferred that embedding the secret data through the ROI by considering the image’s statistical features will assist in obtaining the required results. Soft computing tools: Determining suitable locations for the embedding process has an essential role in embedding the secret data. The determination of such embedding locations is performed based on applying soft computing tools. Applying different optimization algorithms, such as neural networks, can assist in embedding the secret data into the host image in a way that increases the embedded payload, innocuousness, and stego-image quality. Enhancing the secret data’s security: Using the encrypted form of the secret data assists in improving the security. Such techniques as the DES and RSA are applied to acquire an encrypted version of the secret data to be hidden in the cover image. Selecting the most effective cover for hiding the data: researchers have previously concentrated on just applying the optimum selection pertaining to the locations of the data embedding in order to acquire an effective image quality. Nonetheless, the findings show that selecting an appropriate cover image maintains the rigidness of a system against any stego attacks while preserving high embedding payload.
In this subsection, a set of recommendations are provided in order to forward the researchers to develop efficient spatial-domain steganography techniques. The compound of steganography with cryptography: the encryption of the secret data prior to embedding it adds as an extra security layer. If the steganographic algorithm could be exposed by a steganalysis attack, then the encryption has to be broken by the attacker so that the secret data could be possibly recovered. The integration of irreversible and reversible techniques: The integration of reversible and irreversible embedding can raise the security and the embedding payload. The same set of pixels are recursively employed by number of different reversible and irreversible techniques where it is hard for an attacker to have the secret data recovered. Hybrid embedding techniques: Multiple embedding techniques can raise the security of the data and can cause confusion with some steganalysis techniques. Additionally, the weaknesses and strengths of the available techniques are exploited for designing a more effective embedding technique. Hybrid embedding techniques might likely represent effective techniques in terms of security and protection. Universal steganography: The study demonstrates that the majority of the available steganographic techniques represent domain and format/type dependents. It is significant that the universal image steganographic techniques are revealed and formed in a way not to rely on the domain or type. Moreover, these techniques offer effective resistances for different attacks. Minimizing the additive noise distortion: Minimizing the distortion resulting from the additive noise can resist modern steganalysis. In general, modern steganalysis attacks compute various distinctive features pertaining to the cover image and stego-image in order to differentiate the images types. At most, such features can be created based on an additive noise exists in the stego-image. Accordingly, challenges for reducing the additive noise in developing new embedding techniques are still in demand. Blind (cover-less and key-less) extraction approaches: Both approaches refer to the capability of recovering the embedded secret data from the stego-image without the need for the cover image or the stego-key. When the original cover image is needed for the extraction procedure, the cover image gets suspicious. In the same context, sending a stego-key might likely be alarming. Consequently, the blind (cover-less and key-less) extraction procedure improves the security of the embedding techniques. Multi-purpose embedding techniques: Many of these techniques are formed in order to achieve a single goal by either acquiring high embedding payload or high imperceptibility. A multi-purpose embedding technique can minimize the method’s complexity and streamline the implementation. In fact, real-time applications acquire these benefits when designing multi-purpose steganography methods. Ideal image steganography techniques must provide high imperceptibility, high embedding payload, and resistance towards statistical steganalysis attacks. However, no any ideal steganography technique in reality. All indicated techniques have merits and demerits, which rely on the adopted algorithm and their applications’ types. Subsequently, the significance of a steganography method is based on the provided application.
In this review paper, a comprehensive survey related to recent spatial-domain embedding techniques are introduced. The difference between information hiding and cryptography is provided. Comparisons among available proposed embedding techniques in the spatial domain are explained based on their merits and demerits according to a graphical and tabular design. Additionally, many different recommendations, which might assist future researchers to proceed further in the spatial-image steganography, are elaborated in this paper. The major challenges pertaining to spatial-domain image steganography are comprised of the followings: (i) Maintaining imperceptibility within an increased level, (ii) Giving an increased security for the hidden secret data, (iii) Providing robust procedures towards many different intruder attacks and (iv) Providing an increased embedding payload. Generally, the majority of spatial-domain steganography techniques are considered more appropriate if high embedding payload is persistently required. However, the most commonly found flaw of spatial-domain steganography is the weak defense against geometric attacks, such as scaling, rotation, and cropping. As per the literature, it is inferred that adaptive embedding techniques are effective, and thus, the research may be directed towards applying adaptive approaches for high quality steganography techniques.