Integrating Genetic and Chaotic Algorithms for Secure Data Hiding: An Enhanced Steganographic Framework

In traditional digital communication, researchers have been analysing data security [1]. Methods for hiding data are implemented to secure information from threats and to sustain data privacy, protection, accuracy, and confidentiality [2, 3]. Data concealment techniques are mainly categorized into three types comprising steganography, digital watermarking, and cryptography. The most widely used data-concealment method for converting raw information into encoded information is cryptography [4, 5]. Raw information is retrieved by decoding the encoded cipher data. Steganography is a process to protect additional information within digital files [6]. The original data and the carrier medium that hides it can be separate digital entities. The embedded data made invisible to an attacker is the core idea of the steganographic method. The presence of secreted information within the communicated protection object is known only to the originator and the receiver; it can be securely sent through a public medium [7].

The digital cover objects are secured through digital watermarking methods. Both a digital signature and a digital watermark provide comparable protection. A digital watermark acts as an embedded marker, for instance, that determines the ownership of a digital object or validates its reliability and originality. Digital watermarking methods are commonly used to regulate data consistency, validate the data sources, and protect the ownership of multimedia files. Protection of multimedia files is connected to the conventional digital watermarking methods. Currently, employing these methods for protecting data with specific features [8] is beginning to increase recognition.

Image steganography is an effective technique to conceal sensitive information within digital images, offering confidential communication [9-13] and data security. Cryptography ensures that the existence of concealed information leftovers unnoticeable to unauthorized parties, thus improving the protection and confidentiality of hidden information. While encryption encodes the message content into an unreadable format to unauthorized individuals, steganography can be integrated with encryption to offer a further level of protection [14]. Even in the case of an interception occurring in an encrypted message, steganography can secure the information from unauthorized access, revealing the existence of concealed data. Unlike encryption, which generates signals that alert attackers to the occurrence of concealed data, steganography remains difficult to identify without specialized knowledge or advanced evaluation methods. Image steganography consists of multiple approaches, each using a different technique to conceal data effectively.

In this part, a literature survey is provided for representative papers on IWT, chaotic-map-based methods, GA and other optimization methods used in image steganography and watermarking.

Muhuri et al. [15] designed an image steganography scheme based on IWT and Particle Swarm Optimization (PSO) to find the proper locations for embedding secret messages. According to the described scheme, PSO explores the existing search space and identifies the proper embedding regions, thereby increasing the robustness and scalability of the data-hiding process. Another optimized steganographic approach that employs IWT was suggested by Sabeti et al. [16]. In particular, the authors used GA for identifying the embedding regions in a flexible manner. In particular, GA represents the candidate solutions in the form of a genetic structure and performs the selection procedure iteratively. As a result, the steganography scheme becomes more adaptive and scalable and thus ensures better robustness of information concealment.

Ambika and Biradar [17] developed an IWT-based steganographic scheme in combination with fractal encryption. In this case, image blocks are transformed one by one using their respective fractal limitations in command to increase the refuge of the hidden information and robustness to any unauthorized access and analysis of the steganography system. An approach that uses an unsupervised neural network for finding suitable cover images and guiding the process of choosing the embedding region was introduced by Kumar and Hussaini [18]. The suggested approach also uses chaotic structures and transformation techniques, including IWT.

Pramanik [19] offered an IWT-based steganographic scheme optimized via GA to make the embedding information more resilient. This scheme uses the edge intensity criterion calculated from the neighbourhood information in order to recognize the appropriate constants for the data implanting. Also, Pramanik et al. [20] presented a steganographic scheme that uses bi-orthogonal wavelet transformation and GA. In particular, the cover duplicate is decayed into various resolution levels and spectral sub-bands using bi-orthogonal ripple transform, and thus creates suitable places for inserting the secret data. The decomposition technique allows to keep the hidden information visually imperceptible and limits the quantity of alteration extra to the original image.

Govindasamy [21] studied a coverless image steganography scheme, where the secret information is directly related to the image without requiring any additional cover image. In this way, the proposed concept offers another direction for data hiding and can allow the image to represent the message without making any changes to its visual representation. Particular attention in this paper is paid to the possibility of increasing the information transmission capability of coverless steganography using Haar IWT.

Li et al. [22] introduced a GAN-based spatial steganographic framework. In this framework, the generator learns complex probability distributions of images and creates a detailed probability map from the deep-level representations of the images. This probability map then guides the embedding process to the more suitable regions, and therefore increases the capability to hide information using the steganographic system. On the contrary [23], the robustness of the steganographic systems to channel distortions was improved. The authors analysed the sources of the channel errors and divided them into steganography-dependent and steganography-independent ones. The proposed approach improves the fighting to JPEG recompression and improvement sifting, thereby addressing practical channel distortions that were insufficiently considered in traditional steganographic techniques.

Pan et al. [24] use double-matrix decomposition in connection with multi-region coverage for image steganography. In this way, the proposed framework aims to increase the recovery ability of stego-images under external disturbances and to increase the security of the implanting course at the similar time. In particular, the chosen data-hiding positions in the wavelet domain are processed via Hessenberg matrix decomposition, thereby increasing the robustness of the steganographic system. Al-Shaarani and Gutub [25] studied two secret sharing schemes, namely, counting-based and derivative-matrix-based schemes, in conjunction with image steganography. In particular, the authors took into consideration both the Least Significant Bit (LSB) and Discrete Wavelet Transform (DWT)-based methods and aimed to combine cryptographic and steganographic schemes.

Peter et al. [26] suggested a data-hiding framework supporting the encryption and decryption processes in connection with the quick response mechanism and shifting-based mechanism. The combination of these two mechanisms is an alternative to the traditional way of protecting and processing concealed information. Setiadi [27] proposed the hybrid edge-detection technique was suggested in order to expand the regions with edges in the cover images. This technique works on the three Maximum Significant Bits (MSBs) of the iWizard pels and aims to increase the number of regions with edges that can be used for data embedding. As a rule, the regions with edges can endure more changes than the smooth ones, and therefore their use can potentially increase EC while keeping good visual quality.

The Ballot transform was used by Hossain et al. [28] in order to generate integer polynomial sequences serving as the coefficients for the non-overlapping groups of m pixels in the shelter copy. GA was incorporated to increase the quality of the generated stage-double without the need for exhaustive evaluation of all the k-bit combinations. Xu et al. [29] introduced the reversible image steganography method called RIIS was developed. In particular, it uses a provisional regulating flow architecture to perfect the supply of residual high-occurrence machineries and contextual information from the host image. Moreover, distortion-guided modulation is performed for the flow-based blocks in order to create a generalized model capable of meeting different embedding requirements. Mandal et al. [30] and Kaur et al. [31] presented the comprehensive investigations and extensive experimental evaluations of digital image steganography.

Wang et al. [32] aimed at the use content similarity to increase the security of JPEG steganography. In particular, the proposed approach exploits the similarity properties of Discrete Cosine Transform (DCT) blocks and uses 64 parallel channels for constructing the effective distortion metric. Then, the distortion calculation is optimized created on the content properties of the twin in instruction to improve the concealment of the fixed information.

In the field of digital watermarking, Soualmi and Laouamer [33] suggested a copyright protection framework combining Speeded-Up Robust Features (SURF), DCT, and the Imperialist Competitive Algorithm (ICA). The main goal of this framework is to protect digital images against any unauthorized modification and reproduction. ICA is used to find the suitable embedding region for the watermarks and to reduce the likelihood of selecting improper regions that would decrease watermarking accuracy and decrease image quality. Soualmi et al. [34] developed the blind color-image watermarking scheme using the A* algorithm, DCT, and a chaotic sequence in order to solve the problems of invisibility, robustness, computational complexity, blind extraction and security. First, the swarm copy is changed to the DCT field in order to obtain frequency coefficients. Then, the A* algorithm and the Arnold chaotic map are used to optimally and randomly permute the sequence of image blocks according to the obtained DCT coefficients. Finally, the embedded information is extracted using the data-hiding scheme.

Thus, the surveyed literature demonstrates that the combination of transform-domain techniques, chaotic maps, evolutionary optimization techniques, and intelligent learning models has improved the recital of duplicate coding and watermarking systems significantly. However, the present approaches emphasize the optimization of individual aspects, such as EC, imperceptibility, robustness, etc., without paying enough attention to simultaneous consideration of these competing aspects. Also, some methods use fixed or manually selected embedding regions, while other methods have certain limitations when keeping image quality at high payload levels. These factors show the need for an adaptive and multi-objective optimization scheme that will allow selecting the secure embedding location, distributing the payload properly, and preserving the perceptual quality of stego-images. The current research aims to provide such an approach by combining chaotic pseudo-random embedding with IWT-based decomposition and GA-based multi-objective optimization.

Table 1 shows a relative appraisal of state-of-the-sculpture image steganography methods with their respective trial results. The efficacy of the future steganography methodologies is measured with the help of some popularly used parameters for evaluating the perceptual quality, structural similarity, degree of distortion, visual fidelity, and security from attack on image-based systems. Some of the most commonly used evaluation metrics are SSIM, PSNR, VIF, Correlation Coefficient (CC), mean squared error (MSE), Quality Index (QI), Image Fidelity (IF), Average Difference (AD), Unified Average Changing Intensity (UACI), and Number of Pixels Change Rate (NPCR).

Table 1. Summary of the state of the performance of techniques

Overall, these studies effectively facilitate the advancement of image encryption algorithms by offering a variety of approaches to enhance protection, efficiency, and durability for securing confidential image data in a variety of fields of applications.

The research uses a standard desktop PC with an enhanced added feature and equipped with a processor operating at 2.3 GHz and able to handle tasks efficiently. The PC is also equipped with memory of 16 GB RAM, providing temporary storage for prompt access. The publicly accessible databases contain standard images comprising Baboon, Barbara, Cameraman, Lena, and Peppers, are incorporated in the data collection.

5.1 Dataset description

Experimental evaluation was done by means of a dataset that included 1000 benchmark grayscale images obtained from open-source image processing repositories such as the USC-SIPI Copy Record and others standard copy-dispensation benchmarks. This dataset included various categories of images including natural images, human faces, urban constructions, plants, texture images, animals, and artificial objects.

The resolution statistics of the dataset were as follows:

• 200 images of resolution 64 × 64 pixels

• 250 imageries of resolve 128 × 128 pixels

• 300 imageries of resolve 256 × 256 pixels

• 250 imageries of resolve 512 × 512 pixels

For the sake of visualization, eight sample images (Lena, Ape, Peppercorns, Love apple, Fortress, Greeneries, Cameraman, and Chute) of resolution 256 × 256 pixels were chosen for further results presentation.

In order to assess payload capacity, a random set of 500 images was chosen from the full dataset since the payload capacity measurement required repeating payload embedding at various payload levels. Image quality measures (PSNR, MSE, SSIM, VIF, histogram evaluation, and robustness evaluation) were performed on the whole set of 1000 images.

The payload levels were 1.0-4.0 bits per pixel (BPP) and the average cargo level reached 3.67 BPP. The performance parameters used in this study represent the average performance on the tested dataset.

5.2 Performance metrics

5.2.1 Peak signal-to-noise ratio

The usually used measured in copy and video dispensation is the PSNR, which is used to appraise the excellence of a recreated or trodden image comparative to the innovative image. The PSNR is calculated using Eq. (4):

$\operatorname{PSNR}=10 \log _{10}\left(\frac{L^2}{\mathrm{MSE}}\right)$                      (4)

where, $L$ denotes the supreme imaginable pixel intensity cost of the image (typically 255 for an 8-bit twin), and MSE denotes the 'mean squared error', intended as the regular formed alteration among the conforming pels of the unique and reconstructed pictures.

The MSE is defined as:

MSE $=\frac{1}{M * N} \sum_{i=0}^M \sum_{j=0}^N[I(i, j)-\hat{I}(i, j)]^2$                     (5)

where, $M$ and $N$ represent the quantity of rackets and pillars in the image, individually, $I(i, j)$ means the original image pel at location $(i, j)$, and $\hat{I}(i, j)$ denotes the corresponding reconstructed or stego image pixel.

5.2.2 Mean squared error

MSE is a basic measure used extensively in image processing to determine how different an original image and a processed or reconstructed one are from each other. It is the MSE between corresponding pixels in the two images. The inferior the MSE worth, the lesser will be the deviation of the processed image from the original one, hence the lower level of distortion in it. The formula for MSE is:

$M S E=\frac{1}{n} \sum_{i m}^n\left(x_t-y_t^2\right)$                    (6)

where, n is the entire quantity of rudiments in the datasets (e.g., pixels in a double), x_i and y_i are the respective basics in the source and the reproduced signal, individually.

5.2.3 Structural similarity index measure

SSIM is a double quality valuation metric that aims at measuring the physical comparation among a position double and an image that is subjected to some transformation. Unlike error measures like MSE, the SSIM metric takes into account the nature of human vision and hence is based on three different elements which include: luminance, contrast, and structure.

The variety of morals for SSIM is −1 to 1; the developed the value of SSIM, the higher the structural comparation among the images. SSIM for identical images is very close to 1. SSIM is computed as:

$\operatorname{SSIM}(x, y)=\left[\frac{2 \mu_x \mu_y+C_1}{\mu_x^2+\mu_y^2+C_1}\right]\left[\frac{2 \sigma_x \sigma_y+C_2}{\sigma_x^2+\sigma_y^2+C_2}\right]\left[\frac{\sigma_{x y}+C_3}{\sigma_x \sigma_y+C_3}\right]$                (7)

where, $x$ and $y$ denote the reference and processed images, respectively. The parameters $\mu_x$ and $\mu_y$ represent the mean intensities of the two images, while $\sigma_x$ and $\sigma_y$ denote their corresponding typical deviations. The term $\sigma_{x y}$ represents the covariance between the pixel intensities of $x$ and $y$. The constants $C_1, C_2$, and $C_3$ are stabilization parameters introduced to prevent numerical instability and avoid division by zero during computation. By jointly considering luminance, contrast, and structural consistency, SSIM provides an effective measure for evaluating the extent to which the proposed steganographic technique preserves the visual characteristics of the original protection twin.

5.2.4 Visual information fidelity

VIF metric is utilized to evaluate the quality of images or videos based on human visual observations. In which, how effectively the perceptual data in a degraded image matches with an original (source) image is computed by the way of VIF. The human visual perceptual factors of the VIF metric include perceptual contrast sensitivity, light intensity, and spatial frequencies, thus enhancing a more visually significant quality metric. The evaluation of the resolution of rebuilt or degraded images is performed by VIF, which considers the perceptual data as more essential to human viewers. The purpose of VIF is to eliminate the constraints of established pixel-based metrics comprise MSE and PSNR, that non match with human interpretation is given in Table 3. Eq. (8) represents the formula for calculating VIF, which differentiates the structural resemblances between the source and degraded images at multiple scales and orientations:

$V I F=\frac{1}{N} \sum_{n=1}^N \frac{1}{\sigma_n^2} \rho_n\left(\frac{2 \mu_x \mu_y+c_1}{\mu_x^2+\mu_y^2+c_1}\right)$                (8)

where, N is the number of scales, σ_n is the typical eccentricity of the Gaussian filter at ruler n, ρ_n is the CC amongst the source and degraded imageries at gauge n, μ_x and μ_y are the mean morals of the source and degraded descriptions, individually. C₁ is an endless added to avert partition by zero.

Table 3. Statistical summary across 1000 test images

5.3 Results and discussion

The proposed method has been validated by utilizing approximately 1000 test images of multiple dimensions, comprising 64 × 64, 128 × 128, 256 × 256, and 512 × 512. Among these, the results of 8 important test images of dimensions 256 × 256 are highlighted. The 8 important test images shown in this method include Lena, Baboon, Peppers, Tomato, Castle, Leaves, Cameraman, and Parachute. Furthermore, the descriptions with various cargoes were confirmed and the recital was assessed through the system of measurement provided in Section 4.1. Initially, Table 4 shows the fair comparison setting which is applied to evaluate benchmark. The proposed approach is compared with those proposed in these studies [15, 16, 20] according to the criteria of data set, payload, image resolution, and evaluation measures. As the payloads used in these previous studies are different, this comparison can only give us an indicative estimation of performance. The fact that the data sets, payloads, embedding schemes, and experimental settings do not standardize between the different research studies means that this comparison can only serve as an indicator.

In order to analyze the presentation of the system, it was likened with some of the latest and broadly secondhand steganographic methods described in the poetry. Nevertheless, it is important to understand here that the results used for benchmarking purposes are from previously published studies where different sets of data, payloads, resolution, and other evaluation parameters can be considered. Despite the best efforts to compare the techniques based on the most similar parameters available in the literature, it was not possible to completely standardize all the parameters used for evaluation purposes. The above comparison thus serves the purpose of a performance comparison and not a controlled and balanced one.

The PSNR and MSE morals for the planned exertion are represented in Table 5.

Table 4. Fair comparison setup used for benchmark evaluation

Table 5. Peak signal-to-noise ratio (PSNR) and mean squared error (MSE) values for the proposed scheme

Furthermore, the mean PSNR and MSE morals of the developed method were associated with the conventional state of art approaches. The comparison histogram analysis is represented in Figure 2. In simple, the PSNR measures the difference in pixel intensity values and dimensions between the degraded and source images. Higher PSNR values indicate minimal information loss or distortion across the encryption and decryption phases, thereby offering an improved quality reconstructed image.

Figure 2. Histogram analysis of unique images and stego images of size 256 × 256

The PSNR measurements in Figure 3 indicate robustness of the decrypted images against cropping threats. This study evaluates the performance with respect to Pramanik [19], Peter et al. [26], and Setiadi [27]. The experimental findings indicate that an encrypted image is reproduced successfully even with pixel occlusion as high as 60%. In simple, our proposed encryption method offers robustness against cropping threats, ensuring that the image is successfully recovered even with significant data loss.

Figure 3. Comparison in terms of peak signal-to-noise ratio (PSNR)

Table 6. Structural similarity index measure (SSIM) and visual information fidelity (VIF) values for the planned scheme

The SSIM and VIF values for the planned work are represented in Table 6.

Moreover, the SSIM price in the suggested method is substantially closer to 1 compared to the single LSB-based steganography approach, showing that the embedded message in the stego images remains closely resembles the original image (Figures 4–6). In the case of file size, the proposed approach offers a stego image with minimized file sizes compared to the LSB-based method, achieving approximately 15%. Moreover, the suggested method and the single LSB possess different error rates of approximately 41-44%, as calculated using an MSE-reliant statistical framework. Table 4 represents a summary of the findings obtained from the dataset.

Figure 4. Contrast in terms of mean squared error (MSE)

Figure 5. Comparison in terms of structural similarity index measure (SSIM)

Figure 6. Comparison in terms of visual information fidelity (VIF)

Table 7. Unseen image volume of the planned scheme

The embedded data capability is calculated by testing over 500 images, thereby establishing the average embedding capability of 3.67 BPP. Table 7 represents the freight volume of the depicted taster imageries.

For evaluating the robustness of our encryption approach against clamor-based threats, we intentionally implemented three distinct noise masses 0.005, 0.05, and 0.1, into the encrypted images: Salt and Pepper. Next, we calculated the PSNR by comparing the original images with the conforming deciphered images. Table 8 represents the PSNR values, which indicate the robustness of the decoded descriptions against the noise threats. Additionally, the present encryption approach incorporates the insights of Pramanik [19], Peter et al. [26], and Setiadi et al. [27]. The outcome of the experiment shows that the decrypted images preserve strong recognition despite the noise interference from the existence of Salt and Pepper distortions.

Table 8. Robustness against noise attack

Table 9 represents the evaluation between [35, 36] and the planned technique on the MSB even. The projected process achieves a developed volume than other means, as the table shown. The location map of the proposed approach reduces the capacity cost for correcting prediction errors, and the interpolation procedures accomplish higher prediction accuracy than [35]. The suggested method allows the use of additional pixels as carrier pixels compared to the study [36].

Table 9. Comparison of the embedding rate between current solutions using the MBS plane

Table 10. Comparison of peak signal-to-noise ratio (PSNR) between current solutions using the MBS plane

Table 10 represents the evaluation between the conventional approaches and the proposed approach on the MSB smooth, as determined by PSNR. In all three devices, the approximated images are generated by predicting and recreating every MSB as all three approaches implement the MSB smooth for data inserting operations. This confirms that PSNR values of the approached imageries are stable by the various inserting rates. The PSNR value of the future approach PSNR is lower than [36] but higher than in the study [35]. Even though the proposed approach implements the identical pixels as the study [35], the reproduced image achieves a higher quality because of its enhanced prediction efficiency. In contrast, the study [36] implements a considerably lower number of pixels than the planned method, thereby reducing the number of MSBs to be recreated and resulting in reduced alteration; however, this results in decreased capacity.

5.4 Security evaluation under tested conditions

Resistance to Statistical Analysis: The randomness is introduced into the embedding phases through GA and logistic maps, thereby complicating challenges to determine the occurrence of unseen data using numerical evaluation unaided. The pel morals distribution in the stage twin is carefully similar to the shelter twin, thereby providing the challenge to distinguish between them. A mean pixel intensity analysis and computed standard deviation are implemented on all the images and the outcomes of the images are illustrated in this study. The mean pixel intensity and standard deviation are widely used statistical measures to depict the delivery of pixel morals in a double. The results are represented in Table 11 below. The present security assessment is limited to histogram analysis, statistical similarity measurements, key-space analysis, and noise-based robustness testing. Advanced steganalysis techniques such as RS analysis, Chi-square attacks, Spatial Rich Models (SRM), and deep-learning-based detectors, including XuNet, were not evaluated in the current study. Consequently, the reported security observations should be interpreted as resistance under the tested experimental conditions rather than proof of comprehensive steganographic security.

Table 11. Statistical analysis on the proposed scheme – mean pixel intensity and standard deviation

(1) Effective Main Planetary Complexity

One of the crucial influences contributing to the refuge of a steganographic framework is the complexity of its key space that includes all possible combinations of embedding parameters. The key space of the proposed framework involves multiple parameters of both GA (crossover probability, mutation rate, selection mechanisms) and logistic map (initial condition and control parameter). The use of multiple parameters greatly increases the key space that makes exhaustive search attacks unfeasible in case when an adversary does not know the correct secret configuration.

(2) Sensitivity of the Logistic Chaotic Map

Chaotic maps usually demonstrate a great sensitivity to their initial values; a minor alteration in the initial state chiefs to a drastically different chaotic sequence. Such a characteristic provides extra security to the proposed steganographic framework since the adversary should possess the exact initial value and control parameter to reconstruct the embedding sequence. Thus, a slight error in the key configuration results in totally different sequence of embedding positions, which makes reconstruction of the concealed information very complicated.

(3) Steganalysis and Attacks Resistance

The proposed steganographic framework can withstand various attacks and steganographic analysis methods (LSB-based, histogram-based, noise attacks). Firstly, a combination of evolutionary optimization and pseudo-random embedding sequence based on a chaotic map decreases the predictability of the implanting course. Secondly, the optimization of inserting places makes the statistical and visual modifications caused by the embedding process almost imperceptible.

(4) Optimizing Payload Capacity

A major advantage of the GA-based embedding strategy is the ability to optimize the process of determining the optimal locations and configurations for hiding the secret information. The adaptive optimization approach takes into account the properties of the image and selects regions that can hold the largest amount of information with minimal visual degradation. Therefore, it allows achieving a reasonable cooperation among the payload volume and the inaudibility of the embedding process.

(5) Resisting Recognized-Shelter and Identified-Stage Attacks

In case of recognized-shelter and recognized-stego doses the adversary possesses both the innovative copy and the stego-twin. It makes possible to find the difference amid the two descriptions and detect the entrenched statistics. The sanctuary of the planned framework in such a case relies on minimization of statistical and perceptual difference between the images and unpredictability of the embedding process. GA optimization of embedding configurations minimizes the distortion while logistic embedding distribution is pseudo-random.

(6) Key Space Sensitivity and Security

The important space of the planned steganographic agenda includes the parameters that control the optimization (GA crossover probability, mutation rate, selection strategy) and embedding process (initial condition and regulator stricture of the logistic map). The refuge of the system relies on the sensitivity of the inserting course to the key space parameters; even a trivial alteration in one of the parameters chiefs to an incorrect embedding configuration/sequence and fails the process of reconstruction.