This section evaluates the performance of the proposed steganography technique against ACO-LSB. Parameters were analyzed to evaluate the quality and effectiveness of the hidden steganographic image [20]. Capacity: Number of bits hidden in the cover image. The average difference in pixels between the cover image and the steganographic image to calculate the quality of the steganographic image through the Peak Signal to Noise Ratio (PSNR). A higher PSNR value ensures that the quality of the steganographic image is better [21]. The PSNR value is calculated based on the mean square error, which calculates the squared difference between the steganography and the cover images. The more the error decreases, the lower the Mean Squared Error value. Following are the equations for calculating PSNR and Mean Squared Error (MSE): (1)MSE=1W×H∑i−1W∑J=1H(Xij−yij)2 (2)PSNR=10log⁡2552MSE

W: Width of the image.

H: Height of the image.

x_ij: Pixel value at position (i, j) in the original image.

y_ij: Pixel value at position (i, j) in the distorted or reconstructed image.

The equation in the figure is the sum of the Mean Squared Error (MSE) and the Peak Signal-to-Noise Ratio (PSNR). These two are major functions in major image processing-based software used to determine the quality of an image. MSE is estimated by computing the average of the squared differences between the original image’s pixel values and the particular pixel values in the distorted image. It gives an average error of the two pictures. An MSE value closer to zero means the original and distorted images are identical. Maximum gray level of the image. For an 8 bits image, gray levels are between 0 and 255. PSNR is the Peak Signal-to-Noise Ratio measurement system. It is quantified in decibels. The higher the value of PSNR, the better is the reconstructed image in relation to the original image; therefore, quality will be implied. The relation of PSNR is inversely proportional to MSE. As the MSE falls, the PSNR rises. An average of squared differences between the two images—original and the reconstructed one. Lower values are better values. PSNR is a measure of the quality of the reconstructed image with respect to the original. The higher the value, the better, and this metric is, hence, more interpretable for human perception because it takes into account the dynamic range of pixel values.

The given metrics together are pretty much used for the evaluation of methods of compression and reconstruction of images.

Where x_ij and y_ij denote the pixel values for the cover and steganographic images, respectively, while “W × H” represents the resolution of the cover image. An outstanding indicator of the reception of photographs and recordings is SSIM. Image degradation can be detected by examining the observed changes in the structural image data [22].

To reduce the possibility of an enemy attack, the hidden message is placed inside a single-colored image and sent across a communication channel. The embedding procedure should produce a steganographic image that meets standards for visual imperceptibility and PSNR value, reducing the chances of discovery [23]. During the embedding process of creating a steganographic image, a checksum is added, and another checksum is calculated during the extraction phase. By comparing the two checksums, we can determine if any alterations have occurred. If the steganography has not been compromised, the recovered secret file should have identical contents and format to the original secret file [24]. Subsequently, launching an attack with the aim of creating steganographic changes in images that need to be identified throughout the extraction procedure. Extract the sensitive message and verify its integrity by comparing checksums [25]. Identify the altered bytes in the hidden image by comparing the data bit pair of each byte with its corresponding decoy bit pair in case of detecting an intrusion.

To address the concern regarding the limited evaluation set, we have expanded our evaluation to include a diverse set of images. The new dataset includes images with varying levels of detail, textures, and contrasts to ensure the robustness and effectiveness of the proposed ACO-LSB method across different scenarios. Below is a summary in Table 3 of the expanded evaluation.

The identification of changed bytes involves a comparison mechanism designed to ensure the integrity and accuracy of the embedded data. Dividing the secret data to bytes, each byte has been divided into four pairs of bits. Each of these pairs of bits were embedded in the least significant bits of the cover image pixels. Taken parallel with these data bit pairs are decoy bit pairs that are created and embedded together or in a specific pattern within the same block of an image. In the extraction phase, data bit pairs with corresponding decoy bit pairs are extracted from a cover image. The retrieved data bit pairs are compared with the indicated decoy bit to verify the integrity of every byte of the extracted secret data. In the case of a mismatch of a data bit pair with its corresponding decoy bit pair, this could either be an altered byte or an erroneously built one. This bit difference is then used for error-handling routines by signaling corrupted bytes and then further actions of error correction could entail re-embedding the data or using the error detection and correction codes.

The grayscale image cover conceals covert multimedia files and misleading data to deceive attackers into believing they are authentic. 50% of the steganographic image will be allocated for hiding data. 25% will be allocated for concealing the secret message, while the remaining 25% will be designated for storing dummy data for integrity verification. The steganographic images will have their right half-bytes altered to include two bits of counterfeit data and two bits of the true secret data, while the left half-bytes will remain unaltered to minimize distortion. When embedding, the concealed multimedia file will be seen as a series of bytes, irrespective of its initial format [26]. The grayscale cover conceals a multimedia file that will be segmented into four vertical sections, each comprising two bits. Steganography in such a case will deal with 2-bit blocks of the secret message instead of 2 LSB. We added an explanation since some of the reviewers indicated that the ACO algorithm and its integration with LSB are not explained in enough detail for readers who have limited or no experience with these techniques to follow. This new section presents an overview of the working of the ACO algorithm, explaining each major step in the process from initialization to ant path construction, pheromone updates, and finally termination. The two-bit decoy data will be exact complements of the secret fragment. These are embedded in the third and fourth least significant bits of every byte. The PSNR values of the cover and steganographic images should match those produced by popular image comparison applications such as Image Magic. The greatest hiding capacity of a grayscale cover utilizing 2-LSB to hide a secret image is determined by the formula HC = Width × Height/4. An image with dimensions of 512 × 512 pixels and a file size of 262,144 bytes can store an encoded quantity of 65,536 bytes (64 kilobytes). The cover image divides every byte pixel of the hidden multimedia file into four two-bit pairs. The bit pair is encoded in the two least significant bits (LSB) of bytes in the cover picture, as seen in Fig. 3.

The grayscale cover has data inserted only in the fourth bit of the right half-byte of each byte, with each pixel represented by a single byte (8 bits). The decoy data is put in 2 MSB (Most Significant Bit) of the right half-byte (q2), while the confidential message fragments will be saved in the two least significant bits (q1). as shown in Fig. 4.

During the extraction and verification procedure, any modifications made to the hidden message are identified using two distinct strategies: To conceal the confidential data, the right half-byte (4-LSB) is initially utilized in pair comparison. The two-bit fragments (2-LSB) on the right conceal the true secret information, whereas the two-bit fragments on the left contain the counterfeit data, which is the inverse of the two-bit fragments of the secret information. After the decoy has been re-inverted, the data bit pair and decoy pair of extracted bytes from the secret message are compared during verification. When the byte remains unchanged, there should be correspondence between the two pairings. When byte pairs fail to match, the altered byte is identified. Through this procedure, a list of the locations of modified bytes will be produced. Secondly, the proposed method employs a checksum comparison to detect any alterations that may have been performed on the confidential information. Certain mutations that are undetectable via pair comparison can be identified via checksum comparison of the embedding checksum and the extracting checksum values. The processing of confidential data is depicted in Fig. 5, wherein every byte is divided into pairs of bits, irrespective of the file format of the data. The secret bit pairings are stored in the accessible channel’s least significant bit (LSB) pairs. The MATLAB checksum function uses the MD5 (MessageDigest Algorithm 5) checksum algorithm to generate a checksum for the secret file. For comparison at a later time, the checksum of the file will be concealed within the steganography file as part of the extraction or verification procedure. Additionally, the value of the checksum may be documented in a file for potential future use by the extract-verify module. The embedding procedure utilizes the Extract-Verify main and Stego Bytes methods of algorithms as shown in Figs. 5 and 6.

The division of the stego image into blocks is implemented to achieve various objectives, supported by both theoretical and experimental foundations. Theoretically, dividing the image into smaller blocks enhances manageability and control over the embedding process, facilitating the precise optimization of embedding positions using the Ant Colony Optimization (ACO) algorithm. This approach aligns with the divide-and-conquer strategy, which is well-established in computer science for handling complex tasks. Furthermore, the distribution of hidden data across multiple blocks improves security by reducing the likelihood of detectable patterns or statistical anomalies, thus enhancing resistance to steganalysis. This method has, hence, had significant benefits when tested experimentally against other methods. Simulations conducted on more than 1000 open-source grayscale images demonstrate that the block-based approach can increase embedding capacity by as much as 30% compared to traditional methods, with the quality of the image kept very high at an average PSNR of about 40.5 dB and SSIM of 0.98. In addition, this division into blocks improved the detection resistance by reducing the rate of detection by 20%. These results prove the efficiency of dividing by blocks to achieve improved embedding capacity, image quality, and security with a strong basis for its implementation in the proposed ACO-LSB steganographic method.

The proposed approach has been evaluated utilizing the BOSSbase1.01 dataset, which consists of 10,000 8-bit PGM (Portable Graymap Image) grayscale images. BOSSbase1.01 is one of the most reputed datasets related to digital image forensics and steganography research. This dataset was created to compare schemes of steganography and steganalysis, and hence it contains a huge volume of images, mostly gray-scale and with a 512 × 512 pixel resolution, in lossless formats like PNG (Portable Network Graphics) to avoid compression artifacts. It is primarily used to benchmark the embedding of hidden messages within these very images, using different steganographic techniques, and for testing the efficiency of steganalysis methods aimed at detecting the presence of these hidden messages. Therefore, BOSSbase1.01 underlies the bulk of academic research in this area and hence offers a unique possibility for researchers to compare their new methods and technologies in a unique manner. The initial one thousand photographs were chosen from the collection to participate in the endeavor. From the following URL (Uniform Resource Locator), the dataset was accessed on 01 June 2022, https://dde.binghamton.edu/download/, accessed on 21 July 2024. The Digital Data Embedding Lab at Binghamton University has generated the dataset [27]. With its dimensions of 512 pixels across the width and 512 pixels in height, the image weighs 256 KB. Using the spatial domain LSB method, the confidential image (Girl.bmp) was embedded into each of the one thousand images for the study. One thousand images have been selected from the BOSSbase1.01 dataset. The filename and dimensions of the image “Girl. BMP (Bitmap)” utilized in the experiment are 59.4 KB and 142 × 142 pixels, respectively. The source of the image was the USC-SIPI Image Database (2023).

Three modules are used to do the trial work. Four-LSB steganography is used by the embedding module to hide the hidden picture among a large group of 1000 grayscale PGM (Portable GreyMap) cover images. The data for the secret image is stored in the two least important bits of each four-bit pair in the stego byte. The first two bits of the right half-byte of the stego byte copy, flip, and save the secret bit pair as fake data. The secret information’s checksum value is found and then added to the steganographic picture. The attack module changes the steganographic picture by replacing bits from the right half-bytes, which hold the secret data, with values that are chosen at random from 0 to 15. Looking for a way to attack the steganographic image again [28]. A predetermined quantity of attacked bytes is utilized to enable the evaluation of detection precision through a comparison with the true number of attacks. Extract-Verify is a program that extracts a secret image and verifies its integrity concurrently through pairwise comparisons of secret and decoy bits and checksums. This module is responsible for performing the checksum comparison by matching the embedded checksum value with the checksum value generated during the extraction process. The module will also be used to create a catalog of locations in the extracted secret image where altered bytes have been detected [29]. Division of the carrier image and the secret image into four blocks in the process of embedding enhances the effectiveness and efficiency of the steganographic method. Such division is necessary for several reasons, including increasing manageability. A more manageable division would therefore mean finer control over the processes of embedding and extraction, reducing the chance of errors and hence a high accuracy in concealing data. It enhances the security of the scheme because, due to this, the hidden data will now be spread over different parts of the carrier image, complicating the possibility of detection of any pattern or anomaly in the steganalysis techniques for hidden data. Moreover, it ensures optimum utilization of computational resources with small blocks so that the Ant Colony Optimization algorithm can process them fast and explore the potential positions for embedding within each block. This localized optimization helps achieve higher embedding capacities while preserving image quality.

As illustrated in Fig. 7, the experiment entailed embedding the covert image (Girl.bmp) within one thousand grayscale images extracted from the BOSSbase1.01 dataset. The embedding procedure was executed in tripartite succession: initially, a direct decoy was inserted alongside the secret data; subsequently, an inverted decoy was inserted alongside the secret data, and finally, the secret data itself was embedded in the absence of a decoy as shown in Fig. 7.

Table 4 displays the three PSNR values for the first 30 photos in the collection. The mean PSNR value for all 1000 photos is computed and presented in Table 5. The PSNR for embedding without a decoy is much greater than the PSNR of the suggested approach and optimal capacity expansion since the PSNR3 case substituted just 2 bits per byte compared to 4 bits in the recommended cases [30]. The peak signal-to-noise ratio (PNSR) of the increased capacity exceeds the PSNR of the reference method. This indicates that the proposed method offers improved invisibility and a more efficient way to hide confidential information.

The ACO-LSB method focuses on a balance between embedding capacity, image quality, and detection resistance across a broad range of scenarios. While literature [31] may excel in specific conditions, the ACO-LSB method aims to provide robust performance consistently across various types of images and embedding environments. We conducted additional experiments comparing the ACO-LSB method with several recent steganographic techniques published in the latest literature. The metrics were considered: embedding capacity, PSNR, SSIM, and detection resistance. The detection resistance of the ACO-LSB method was particularly notable, reducing the detection rate by 15% compared to the latest methods, underscoring its robustness against steganalysis.

Fig. 8a illustrates a pristine cover image, while Fig. 8b depicts a steganographic image, and Fig. 8c presents an enhanced capacity image. Even after the 4 LSB replacement, no evident differentiation remains between the clean and steganographic images. There is no discernible difference between the stego image with increased capacity and the original stego image. In order to prevent any discrepancies in the images, an alternative embedding technique was implemented: the data decoy pairings were encoded in alternate bytes utilizing 2 LSB. The steganographic image generated via alternate embedding, employing the cover data image Bird.jpg as illustrated in Fig. 8. When compared to the 4 LSB technique, the utmost secret data size is reduced by 50% when alternate embedding is utilized.

Each of the proposed steganographic images was altered by adding 51,200 bytes. The 4 least significant bits of the bytes were replaced with random values between 0 and 15. The Extract-Verify application was used to analyze one thousand modified steganographic images to extract the hidden secret message and detect any integrity breaches. Following the assault, the confidential image that was obtained is shown in Fig. 9. The assault is clearly visible when analyzing the extracted image, which is a prominent example of an altered document [32]. However, in practical scenarios, attacks on papers could potentially go undetected due to variations in format and content.

The detection rate for the initial 30 images was processed using the pair comparison method. A portion of the total assaults on the steganographic images comprised 12,800 bytes, of which one-fourth was devoted to the secret image. A distinct byte was utilized to store each pair of bits of the secret byte. When the pair comparison method was unable to identify the attacks, the checksum comparison method demonstrated efficacy.

When describing the secret information embedding step, the checksum is a crucial component for ensuring the integrity and accuracy of the embedded data. It is a value calculated from the secret data before its embedding, which is later used for verification at the extraction stage. The checksum enables the checking of hidden data for errors or modifications during its transmission or storage. as shown in Table 6 explains a comprehensive evaluation of the proposed ACO-LSB steganographic method.

Extract-Verify uses pair comparison and gives a listing of the locations and contents of the modified bytes. Table 7 shows some locations and contents of modified bytes of the first image in the dataset Girl.bmb 1. Using this list of modified bytes, one can get an idea of the patterns of the attacks that most frequently occur.

The proposed ACO-LSB steganographic method thus has high resistance to steganalysis, and hence, the techniques of detection find it very hard to discover the hidden data from an image. This strength can be attributed to several important features that have been incorporated into the methodology. First, the adaptive searching provided by the Ant Colony Optimization algorithm makes it dynamically select the optimal positions of the pixels for embedding data. This adaptivity of the algorithm reduces the predictability of the embedding patterns and hence makes it harder for steganalysis algorithms to detect the hidden information. Furthermore, the pheromone update mechanism in the ACO algorithm will refine the selection of the embedding position iteratively so that data will be embedded in such a manner that detectable anomalies are reduced to a minimum. All of these readaptations can help retain as much as possible of the natural statistical properties of the cover image, thereby further complicating the efforts of detection.

While having its different strengths, the ACO-LSB technique has some limitations. One major limitation is the additional complexity and computational cost resulting from the use of the ACO algorithm. The processes of embedding and extraction demand costly computational resources; this might be a weakness where resources are limited or when real-time processing is required. Another limitation is that it depends on the characteristics of the cover image. The method’s effectiveness can depend on the uniformity and texture of the cover image. Scenarios with low texture or uniform areas in the image provide fewer optimal embedding positions, hence probably reducing method capacity and security. Also, the performance of the technique under hostile conditions like huge compression of the stego-image or high noise levels is not fully known. These conditions could also affect the ability to embed, and the precision of data extraction, and most likely cause challenges in real-world applications regarding the reliability of the method.