An Improved Steganographic Scheme Using the Contour Principle to Ensure the Privacy of Medical Data on Digital Images

1  Introduction

One of the classic demands of the contemporary world of the internet with digital communication is the term “secure communication.” The method of hiding secret/confidential Content in a cover image is the theme of the phrase “digital image steganography” to produce the stego image, which would be transmitted to the receiver over the internet medium. The primary goal of the steganography principle is to protect the secret substance from intruders over the internet medium. New strategies and venues for content concealing have emerged due to the availability of cutting-edge computing system facilities. To achieve the content confidentiality feature over the internet medium, the principles of Steganography and Cryptography are utilized, and Table 1 shows the main differences between the two principles.

While both cryptography and steganography are concerned with the secure transmission of information using covert methods, there is a distinction between the two classes. Cryptography utilizes a key and a transformation to encrypt the data into a disguised form. It will have a key for the decryption of the transformed message back into its original state. In steganography, the Content is disguised into a cover object, which is then transmitted and analyzed by the receiver to acquire a secret content-embedded image known as a stego image. Both the stego and the cover image must be protected against deterioration. Steganography is a technology that hides hidden Content in a trustworthy medium, such as digital photos of grayscale or color, digital films/audio, and then pretends that it exists.

Steganography algorithms interpose a large amount of secret Content into many métiers to hide the confidential Content from attackers. The models’ main conducts are the quality of the cover and stego images and their secrecy. Discrete Cosine Transform (DCT) and Discrete Wavelet Transform (DWT) are two examples of steganography based on spatial and transformed domains (DWT). The goal of steganography principles is to hide a large amount of secret or confidential information on a large amount of media. The level of increasing imperceptibility, payload capacity, robustness against multiple attacks, and security are the main criteria for evaluating steganographic algorithms. Other qualities such as reversibility, encryption, and computational complexity are used in diverse applications to increase the stego object’s security or communication. Steganography can be used in a variety of fields, according to current trends, and is dependent on the type of carrier. Steganography curves have been promoted due to recent technological advancements, and some of the trends include medical photographs, multimedia data, network contents, DNA patterns, and so forth. Most hidden content embodiment strategies use the famous Least significant bit (LSB) replacement strategy or the spatial orbit’s pixel-value differencing scheme. One of the commonly established principles for hidden Content concealing on images is the LSB assignment method, which limits the steganographic models. Because any intruder could extract the secret data from the LSB content of the images. To improve the chaotic level of secret content extraction, authors deputize a rigid or variable length of bits in these approaches; the confidential material has been embedded into a digital image. Only if the secret stream for the embodiment procedure is large in quantity does one of the proficient methods, “Pixel-value differencing,” yield more favorable results. The clustering scheme “K-Means” is based on the theory of vector quantization and aids the data mining process of cluster analysis. The K-Means clustering principle leads the sectionalization of the available ‘n’ reflections into ‘K’ clusters, with all observances processed to arrive at a cluster based on its skinniest mean value.

The suggested model has two stages that deal with the secret content embodiment process. With the help of the secret key value, a contour pattern is provided and constituted to a distinct domain in all layers of the digital color image to traverse the pixels of the image. In the resulting contour of the layers, pixel blocks are generated and classified into ‘K’ separate clusters. The secret data embodiment is performed on the contour clusters using the LSB substitution approach in the next phase. As a result, the suggested approach aims to achieve efficient data privacy via pixel values on all image levels and retain secret Content via digital images. After the embodiment procedure, the freshly created stego image is subjected to communication channel transfer. After successfully detecting contour clusters at the receiver end, a reverse method might be used with the shared stego image to extract the original secret data. The proposed model’s public presentation is compared to the results of many previous approaches. The trials’ findings indicate that the nominated secret content embodiment practice on digital photographs is both graceful and reliable.

The proposed scheme perch comprises the following elements: The literature review is covered in segment 2, and the nominated scheme’s primary idea and capabilities are presented in detail in segment 3. In segment 4, the experimental findings and analysis of the nominated scheme are presented. Finally, the nominated scheme’s conclusion is provided in Section 5.

2  Literature Survey

The process of capturing digital photos has never been easier, thanks to digital cameras. Huang et al. [1] presented an approach for reversible and high-capacity data hiding in high-quality cover images. The authors nominated a model for deciding the pixels over the image by dividing the image into tiles and shifting the histograms of each image tile between its minimum and maximum frequency. Raghupathy et al. [2] demonstrated a new pattern for pixel embodiment on a grayscale image using the bishop tour pattern from the chess game. In this work, the authors thoroughly examined modern steganography concepts and provided an FEC-based double security system based on encrypted image Steganography [3]. The research proposed a model for detecting noise in transmitted images and different types of attacks. Another method proposed by the author’s Prasad et al. [4] for implementing RGB color image steganography and the models is to use overlapping block-based pixel-value differencing methods. On the content concealing process, Sathish Shet et al. [5] proposed a competent paradigm for the design and development of reconfigurable architectures. The LSB and multi-bit based steganography principles were used in work mentioned above. Naqvi et al. [6] describe a methodology for steganography implementation based on the multilayer partiality homographic encryption concept. The paper demonstrates that the model provides secure text content transmission with the cover image through communication channels. Authors Manikandan et al. [7] have presented a lossless steganography system that generally adheres to clustering and contouring principles on grayscale photos. Hua et al. [8] an image encryption scheme using value-differencing transformation and modified zigzag transformation. This scheme encrypts different types of images with high level security. The method uses a partial encryption scheme for confidential Content hiding in medical photographs. Konyar et al. [9] offer a clever medical image-based data-hiding strategy for achieving content secrecy and promoting secure data transmission over the communication medium in this work. The secret content embodiment procedure in the study mentioned above employs Reed Solomon coding in conjunction with the salt and pepper noise principle. Another approach for embodiment proposed by Aziz et al. [10] for achieving protected content embodiment on images. The model described above is concerned with reversible data concealment strategies with a large secret content embedding capacity in images. The author Li et al. [11] proposed a method based on the ideas of reversible data concealment and quick response codes. The method uses a partial encryption scheme for confidential Content hiding in medical photographs. Various steganography methods and appropriate steganalysis schemes that have a high level of applicability in the field of digital forensics have been presented in work stated by the authors Dalal et al. [12], and the same authors [13] have compiled a survey for the efficient implementation of video steganography. The researchers Patel et al. [14] presented a full review of the inquiry of video steganography in the uncompressed and compressed domains, in which several compressed and uncompressed mediums with their advantages and disadvantages in terms of information concealing were presented. The authors Lin et al. [15] demonstrated feature-based steganalysis for Pixel-Value Differencing Steganography, and the model demonstrates the significance of statistical features-oriented embodiment and detection techniques. Authors Kordov et al. [16] have proposed another implementation of the steganography principle based on color photos medium images with random order pixel selection and encrypted text message embedding. The author AbdelRaouf [17] proposed a new approach for hiding secret Content on color images, in which the pixel values for the secret content embodiment are chosen based on the visual color sensitivity of the pixels. Manikandan et al. [18] proposed a new method for achieving concealed content embodiment on sensitive medical images. The mentioned works with DICOM pictures. For pixel pattern identification, several chess-based approaches are used to explore the pixel paths, and experimental observations show that the offered model is effective. Another model proposed by authors Krishnan et al. [19] for data hiding on images works by encrypting the image and then applying the secret material to the appropriate pixel channels, with the image being descrambled after the embodiment process to return to its original form. The experimental results demonstrate the model’s chaotic degree and uniqueness. It might be more powerful for detecting the places where content embodiment has been performed. Those regions must be correctly detected at the receiver end to achieve lossless content extraction. Another model suggested by the authors Yassin et al. [20] for the attainment of information hiding in digital images with the help of wavelets. The above-stated model deals with the content concealment process by following the most significant bit substitution methodology for attaining confidentiality. The authors Sabeti et al. [21] presented an adaptive image steganography method for content concealment through integer wavelet transformation. The model utilizes a genetic algorithm for pixel detection on images to perform the secret data concealment process. The authors, Pilania et al. [22], suggested an ROI-based video steganography scheme in the wavelet domain using the SVD mechanism. The above-stated model offers an efficient Steganographic model for attaining a secret stream hiding in video frames.

3  Proposed System

The proposed model has been designed to effectively combine Geodesic active Contour Algorithms, often known as Snake Algorithms, K-Means Clustering principles, and steganography’s Least Significant Bit (LSB) replacement technique. The proposed novel steganographic model is concerned with layering the digital image into Red, Green, and Blue layers. The system flow begins with the contour generation principle, followed by the process of pixel clustering on picture layers such as Red, Green, and Blue, and finally, the process of identifying pixels within the contour on the previously mentioned layers to attain the secured content transmission over the communication channel partners.

The suggested model works by turning the secret Content from the medical client on an internet medium for transmission into a binary stream that may be used in the embedding process. The conversion would be carried out using the standard ASCII-based stream creation method. It generates a secret binary stream, which is then divided into three parts for the purpose of embedding practice on all of the color image’s available channels or layers. The embodiment process’ confidential material, or secret binary stream, has been divided into three variable length stream arrays: Red Stream, Green Stream, and Blue Stream. The kinship between the stream arrays is:

Red_Stream = Secretstream_count/2, Green_Stream = Secretstream_count/8, and Blue_Stream = (3 × Secretstream_count)/8. Where Secretstream_count is the length of the binary stream of the secret code for the embodiment process.

During the content embodiment process, the user must indicate the coordinate position in the digital image for the development of contours, and the Mapping-Key will determine the relevant cluster for the embodiment process (MK). The cluster for the embodiment process would be based on the MK value, and the secret contents of the stream array would be embedded in all three layers of the image using the LSB replacement technique. K-Means uses the LSB approach to channel the clusters on the image layers, resulting in the stego image, a hidden binary stream-embedded image. The stego image would then be shared with the recipient through the internet medium. The secret stream from the shared stego image is recovered using the exact reverse technique at the receiver end (Medical practitioner) for offering services to the client who requested services/suggestions. The workflow of the proposed model scheme at the sender side is shown in Fig. 1, and the workflow at the receiver end is presented in Fig. 2.

Figure 1: Secret content embodiment process at the sender side (Medical Client)

Figure 2: Secret content extraction process at the receiver side (Medical Practitioner)

Algorithm 1 and Algorithm 2 provide a secret content embedding procedure and its corresponding extraction procedure.

The proposed model holds the pixel clusters for the embedding process with better intelligence using Mapping-Key (MK). On the contours of the image layers, the K-Means clustering principle is imposed, resulting in K-Clusters. In each contour, a minimum of 6 clusters can be found based on the dimension of the input cover image, with each cluster holding 6 pixels for the embedding process. As a result, six clusters are formed in each layer of the cover image ‘K’ for the secret content embodiment procedure.

On a color image, the secret stream embodiment phase works by picking out pixels in each cluster one by one, and the emergence can be completed after the embodiment of Red Stream, Green Stream, and Blue Stream content. The secret stream count on the relevant layers might be used to determine the size of the contour zone on the cover works. If the image’s size is insufficient for the embodiment process, the image will be processed by increasing the size of the contour region on the image layers. Fig. 3 shows a sample of the pixel selection process and alternate region selection and clusters.

Figure 3: Clusters generation process for red layer in cover image

The initial values of the pixels in the clustered regions are adjusted after the secret stream embodiment process on the pixels of the Red, Green, and Blue layers of the image. Table 2 shows the initial cluster values, the secret stream in natch of bits, the secret bits embedded cluster value, and the cluster similarity percentage.

4  Experimental Observations

The proposed LSB-based embedding model with Contour and K-Means clustering strategy was developed in MATLAB R2021a software, input DICOM images with layers (Red, Green, and Blue) are taken from the source “Open Access Medical Image Repositories (aylward.org)” [23] and output snapshots of the processed inputs are presented below. The proposed model works with six contour clusters on each layer of the image, and each cluster holds six-pixel values. So that it is possible to hide 12 bits of data using LSB substitution on each cluster, a maximum of 216 bits of private data would be permitted to hide in an image with dimensions 512 × 512. The proposed model’s resilience has been tested with various dimensional values of the input images (as shown in Fig. 4) and varied dimensions of the contours (as shown in Fig. 5). The following four input color images are considered for processing in the experimental outgrowth: (i) Head (ii) Skull (iii) Chest and (iv) Hand. The color images are used as input cover images, and the secret stream has been inserted in the images, resulting in the stego images shown in Fig. 6.

Figure 4: Clusters generation process for red layer in cover image

Figure 5: Various contour sizes generated on various image dimensions

Figure 6: Input cover images vs. stego images

The metrics Mean Squared Error (MSE), Peak Signal to Noise Ratio (PSNR), Number of Pixels Change Rate (NPCR), and Unified Averaged Changed Intensity (UACI) are determined to assess the compatibility of the stego picture with the corresponding cover image. The process for computing MSE, PSNR, NPCR, and UACI for a N × N cover picture and stego scale image is described in Eqs. (1)–(4).

MSE(colorimage)=1p×q∑i=1p∑j=1q[(CImageRedLayer(i,j)−SImageRedLayer(i,j))2+(CImageGreenLayer(i,j)−SImageGreenLayer(i,j))2+(CImageBlueLayer(i,j)−SImageBlueLayer(i,j)2)] (1)

where CImage and SImage  represents the cover and stego images, respectively. The symbols RedLayer(i, j) , GreenLayer(i, j) and BlueLayer(i, j) denote the pixels (Red, Green, and Blue) in locations (i, j) of the C_Image and S_Image. The parameters p x q denote the image dimensions.

PSNR=10 log10 2552MSE  (2)

where MSE represents Mean Squared Error.

UACI =1p x q∑⁡p,q∣C_Image(i,j)−S_Image(i,j)∣255∗100%, (3)

NPCR=∑⁡p,q D(p, q)p x q,  (4)

D(p, q)=∑k=0n⁡(nk)xkan−k

D(p, q)={1 if C_Image (i, j) ≠S_Image(i, j)|0 if C_Image (i, j) =S_Image(i, j)} (5)

where C_Image and S_Image represent the Cover and Stego images, respectively, the symbols ‘p’ and ‘q’ represent the images’ dimensions. The parameters ‘i’ and ‘j’ represents the pixel locations on the images.

The metrics MSE, PSNR, NPCR, and UACI for different cover and stego image dimensions with varying contour sizes are computed to prove the efficiency of the chosen approach. The MSE and PSNR values for the collection of input cover images with different dimensions are shown in Tables 3–5 shows the results of the NPCR and UACI tests that were performed. Figs. 7 and 8 show the evolution of the nominated method’s MSE and PSNR values on various image sizes. The proposed approach was compared with various existing steganography models [7] for embedding secret Content, which is listed in Table 6, and graphical Content is shown in Fig. 9. According to the findings, the observed experimental values are close to the theoretical ideal values. As a result, the suggested embedding approach can effectively shield the hidden Content from ordinary perception.

Figure 7: MSE of the anticipated scheme with distinct image dimensions

Figure 8: PSNR of the anticipated scheme with distinct image dimensions

Figure 9: Comparison of the nominated scheme with existing outcomes

5  Conclusion

The nominated steganographic model gives an ameliorated scheme using Contours and K-Means clustering on digital images to accomplish secret Content (from a medical client) hiding on all levels of the color image with dimension 512 * 512, according to our provided method and its investigational observations to generate a stego image for transmission. The model generates a stego image with the secret Content, and the novelty of the proposed scheme is evaluated using the metrics Peak Signal Noise Ratio, Number of Pixels Change Rate (NPCR), and Unified Averaged Changed Intensity (UACI). Comparing the proposed model with existing schemes is encouraging to learn that the proposed scheme consistently outperforms existing standard schemes in terms of quality and other security-related activities. The proposed principle is very effective and ideal for information concealment in telemedicine applications. In simple terms, the proposed system satisfies all the requisites necessary for information security-related applications and efficiently performs the desired function. The future path of this study work would be to incorporate the presented model on videos with the deployment of Privacy preservation techniques.

Funding Statement: The authors received no specific funding for this study.

Conflicts of Interest: The authors declare that they have no conflicts of interest to report regarding the present study.