Enhancement on Secure Transmission of DICOM Images with Optimized Chaos Based Encryption and SPIHT-SVD with QIM Based Watermarking

Medical image communications, data storage, and transport have all grown in popularity during the last several decades. Network assaults, denial-of-service, phishing, and man-in-the-middle are among the cyber threats that data encounters during transmission over the internet. Data encryption systems need to be trustworthy and secure in order to guarantee the safety of information while it is being sent. The spatial or frequency domains are often the basis for image encryption (IE). The frequency domain is occupied by IEs like wavelets and Fourier transforms, whereas the spatial domain is occupied by IEs like chaotic maps, DNA encoding, cellular automata, and compressed sensing [1]. Some features of digital images include data redundancy, strong correlation between neighboring pixels, reduced sensitivity to changes in image quality caused by small changes in pixel attributes, and large data storage capacities, among other things. Since they need a great deal of processing power and time to decipher, standard ciphers such as IDEA, AES, DES, RSA, etc., are unfit for use in real-time image encryption. Crypts that are both fast and secure are the only ones that should be used for real-time image encryption. Even with a high level of security, real-time operations would be severely hindered by an inefficient encryption method [2]. Due to its great dynamism, complexity, and sensitivity to control parameters of the maps, chaos-based IE is the most often used system. Two processes, permutation and diffusion, are often used to run IE systems based on chaotic maps.

In order to address the growing need for secure real-time image transmission across wireless networks and the Internet, image encryption techniques have been extensively researched. When dealing with very big images, traditional image encryption algorithms like data encryption standard (DES) struggle due to their poor degree of performance. The challenge of providing quick and very secure image encryption has proven to be intractable, but chaos-based encryption has proposed a novel and effective solution. Image encryption research has increasingly relied on chaotic systems since 1989, when Matthews introduced the chaotic encryption technique. A chaotic encryption system has been the subject of several recent publications [3]. Traditional encryption algorithms and chaotic-based encryption systems are only two examples of the numerous available encryption approaches. But some of these older encryption techniques are complicated, sluggish, and difficult to learn, and unfit for use in real-time applications. However, the benefit of disorder resides in the fact that unapproved users see it as noise, which is a major deterrent. As a result, encryption systems based on chaos provide secure transmission networks that are both quick and simple to build [4].

Image processing, signal processing, electric power systems, optical assessment, neural networks, and many more scientific fields make use of chaos, a stochastic phenomenon that arises in nonlinear and guaranteed systems with the properties of randomness, regularity, ergodicity, and sensitivity to the initial values. Combining chaos with optimization methods for issues improves the efficiency of image encryption [5]. This is because the qualities of chaos are similar to those of swarm intelligence. Chaotic systems, which are ergodic and very sensitive to the secret key—a combination of the system characteristics and beginning conditions—also possess the two crucial features of confusion and diffusion—and so are essential to any successful conventional encryption scheme. Analogue chaotic cryptosystems use discrete dynamical systems, whereas digital chaotic cryptosystems use continuous dynamical systems; these two types of methods are based on the categorization of chaotic systems. Also, the architecture of chaos-based encryption algorithms tends to be somewhat simple. Consequently, many cryptosystems based on chaos have been suggested recently [6].

Compressing an image implies reducing its file size in bytes without drastically lowering its quality. Eliminating duplication is the primary objective of compression. The number of meta-heuristic algorithms has increased in recent years due to the proliferation of optimization issues. These algorithms include genetic, ant colony, particle swarm, fish swarm, Virus colony search (VCS), fire-fly, and many more. One such technique that Professor Pan Wenchao introduced in 2011 called the fruit fly optimization algorithm (FOA) has a basic structure, is straightforward to select parameters for, and converges quickly. Managers, forecasters, planners, etc., in the economic sector rely heavily on FOA. However, since the fly step size was set in the original FOA, it may have sunk into local optima like other optimum evolutionary algorithms. Denoising performance is impacted by the fitness function, which is a critical issue. Evidence has been presented by a number of academics. But the vast majority of them include complete citations. Both the methodology and the reference images are impractical for real-world use [7].

The challenge of decreasing the data needed to depict a digital image is the focus of image compression. First, coding redundancy, which occurs when coding is suboptimal; second, inter-pixel redundancy, which arises from pixel correlations; and third, psycho-visual redundancies, which arise from data that the human visual system ignores, are the three primary types of data redundancies that must be removed in order to achieve compression [8]. By using SPIHT, the grey level image compression technique may be extended to compress colour images as well. The SPIHT method, in relation to lowering thresholds, produces an embedded bit stream of wavelet coefficients. Extremely low image quality is the outcome when an excessive amount of bits in the encoded bit stream indicate insignificance. A confirmation of better reconstructed image quality may be achieved by reordering the bits in the encoded bit stream so that important information is given priority. Because of this, the energy compression of the wavelet coefficients may be used more effectively [9].

When it comes to assuring the validity, integrity, and security of data in the healthcare industry, medical image watermarking is very necessary. Protecting sensitive patient information from being accessed or altered by unauthorised parties is accomplished by the incorporation of watermarks into medical images. This method also makes it possible to verify the authenticity of the image, since any modification would cause the watermark to become disrupted, which would indicate that the image may have been tampered with. In addition, watermarking lends assistance to copyright protection, which guarantees that the ownership of the medical data as well as its point of origin are maintained. In the grand scheme of things, it is an extremely important factor in ensuring the dependability and secrecy of medical images in clinical and research environments. Chaos-based image encryption is crucial in ensuring the security of digital images, because of its remarkable capability of generating intricate and unpredictable patterns. By utilizing the principles of chaos theory, this method guarantees that even the smallest alteration in the encryption key will produce a wholly distinct encrypted image, thereby greatly impeding unauthorized individuals from deciphering the encryption. Chaos-based systems possess a unique ability to effectively safeguard large and sensitive image data in a wide range of applications due to their inherent sensitivity and unpredictability.

When decomposing matrices, SVD and HD are the tools of choice. False positives are an issue with SVD processing, however they can be solved by encrypting the SVD components with chaotic systems. The objective evolution function (oefunc) aids FOA in finding an optimised scaling factor, which enhances efficiency and helps preserve the watermarking model's balance between invisibility and resilience. A methodology for first encrypting images and then watermarking them is worked in this paper. Using the fruit fly optimisation method, Chaotic has optimised its encryption. The Quantisation Index Module (SPHIT) with SVD will be used for embedding.

A combined compression and encryption strategy that executes compression and encryption in one step is a promising solution when taking into account the qualities of image data, image processing technology, security needs, and the rising amount of images. The length of the cipher image is made unknown by the interplay between the compression operation and the encryption procedure. The encryption and watermarking of proposed work are explained in Figure 1.

0401.png

Figure 1. Encryption and watermarking of proposed method

The flowcharts illustrate the processes and procedures involved in image encryption, each of which makes use of a unique set of methods to improve safety. The first chart is an illustration of a technique that utilizes chaotic maps with firefly optimization. From the beginning, an input image is transformed into chaotic maps, which results in the production of sequences that are subsequently concatenated. An image that has been encrypted is the end product, and it is then subjected to further optimization employing firefly methods in order to guarantee its resilience and security. A procedure that combines the SPIHT algorithm with the SVD technique is shown in the second flowchart. The process starts with a cover image, which is then altered using SPIHT in order to ease the production of an effective key. After that, the procedure makes use of SVD for further augmentation, which results in the generation of a stego image that incorporates the information from the original image. After all is said and done, an encrypted image is created by using key generation and embedding procedures. This ensures that the data is hidden in a safe manner.

3.1 Preprocessing

Preprocessing steps in image processing and computer vision sometimes include resizing images. One way to filter out background noise while keeping edges intact is via the Alpha Trimmed Mean Filter technique. The Alpha Trimmed Mean Filter technique is useful for determining the neighborhood's lowest gray-level d/2 value and greatest g(s,t) d/2 value. After removing the lowest and greatest values by d/2, this filter substitutes the average grey level value in the subimage beneath the size m×n adjacency window for each pixel's value. Here is the equation that defines the Alpha Trimmed Mean Filter method:

$f(x, y)=\frac{1}{m n-d} \sum(s, t) \in S x y g(s, t)$                   (1)

The filter kernel size is denoted by mn, the input value ranges from 0 to 9, the pixel intensity value that will be replaced by the filtering result value is g(s,t), and the number of pixel intensities influenced by the filtering process is $\sum(s, t) \in{Sxyg}(s, t)$.

3.2 Chaotic optimized with fruit fly optimization for encryption method

To make algorithms seem more random, statistic distributions like the uniform and Gaussian distributions are used. Given its unpredictability qualities, chaos is an excellent option for producing random data. Algorithms may be able to complete iterative search steps more quickly than traditional stochastic search using normal probability distributions due to the chaotic properties of ergodicity and mixing chaos [5]. With its straightforward operation and well-dynamic unpredictability, logistic mapping—introduced by May in 1976—is the most emblematic chaotic mapping system. One definition of logistic mapping is:

$\begin{gathered}z(t+1)=\mu z(t)(1-z(t)) \\ z \in(0,1), 0<\mu \leq 4\end{gathered}$                     (2)

The iteration number is represented by t and control parameter c decides whether the chaotic variable z stabilises at a constant value. You can't put a value of 0, 0.25, 0.75, 0.5, or 1 into the variable z. When the value of μ is 4, the logistic mapping sequence becomes chaotic. The value of 4 μ = is used in subsequent investigations [5]. Using a 256×256 key for a 256×256 image in chaotic cryptography guarantees that every pixel or block of the image is safely encrypted, providing enough key variety and defense against brute-force assaults. By matching the image dimensions, this key size maximises computing efficiency and security.

A novel global optimisation algorithm called FFOA was developed based on the entrapment behaviour of fruit flies. Using a model inspired by fruit flies, the FFOA approach finds the optimal solution to the optimisation issue. Sensitive olfactory organs let them identify a target's aroma, and then acute visual organs helped them find their way there. The original fruit fly optimisation approach (osphresis foraging, start, population assessment, and vision) [16] comprises these steps.

• Step 1: Let $\ X_a, Y_a$ represent the positions of the fruit fly swarm. Initialize these variables randomly.
• The methodology has population size = 50, total iteration = 500, with a step size of = 0.1.
• Step $2:$ Let $\ X_i=\ X_b+x_r, \ Y_i=\ Y_b+y_r$, where $x_a$ and $y_r$ are the variables for random values. And let the location coordinates be $\ X_b$ and $Y_b$, where $\ X_d, Y_a$ are the initial values of these location coordinates, gives information about distance and random direction of the individual fruit fly.
• Step 3: Let $D_i=\sqrt{X_i^2 + Y_i^2}$ is the distance measured from the origin. $D_i$ is calculated first and then the reciprocal of $\ D_i$ is calculated which is called as the smell concentration $S i=1 /  D_i$.
• Step 4: At individual location of the fruit fly the smell concentration is given as Smell$_i$. Find Smell$_i$ which is a function of $\mathrm{S}_{\mathrm{i}}$, and it is denoted as Smell$_i$ = Function $\left(S_i\right)$. Function $\left(S_i\right)$ is the smell concentration fudgment function.
• Step 5: Find the maximal value of the smell concentration in the swarm of fruit flies, and it is given as [bestSmell, bestIndex] =max (Smell$_i$).
• Step 6: The best values of the smell concentration and the coordinate is recorded, then the swarm of the fruit flies will move towards the final location using vision Smellbest = bestSmell, $\mathrm{X}_{\mathrm{b}}=$ $X(bestIndex), \mathrm{Y}_{\mathrm{b}}=Y(bestIndex)$. 
• Step 7: Perform iterative optimization and repeat Steps 2-5 if the iterative smell concentration is better than its previous value. Otherwise, return to Step 6.

3.3 SPIHT

It is common practice to combine image compression algorithms, such as SPIHT, with watermarking methods to include watermarks into compressed domains. When SPIHT is used to watermarking, it becomes much more resistant and almost undetectable. Image compression is the main use of the SPIHT method. As more bits are received, the image quality increases, and it is very efficient. It also offers progressive image transmission. In contrast to JPEG 2000, the SPIHT method does not aim to minimise memory or bandwidth and is not optimised to examine areas of interest, yet it achieves a high compression rate when N is big. When it comes to lossy image compression, the SPIHT algorithm is often considered to be among the most effective methods [14]. In 1996, Said and Pearlman created the SPIHT image coding technique, which is an improved version of Shapiro's embedded zerotree wavelet (EZW) algorithm [10]. Following an image's application of the wavelet transform, the primary algorithm divides the resulting wavelet decomposed image into meaningful and irrelevant parts using the following function:

$\operatorname{Sn}(T)=\left(\begin{array}{cc}1, & \max _{(i, j) \in T}\left\{\left|c_{(i, j)}\right|\right\} \geq 2^n \\ 0, & \text { Otherwise }\end{array}\right)$                    (3)

The significance of a set of coordinates T is denoted as Sn(T), and the value of the coefficient at coordinate (i,j) is ci,j. The sorting pass and the refining pass are the two stages of the algorithm [14, 15]. The SPIHT technique compresses images efficiently using redundancy. After converting the input image to grayscale, it uses a wavelet transform like the 2D Discrete Wavelet Transform to divide it into frequency bands. Starting with the List of Significant Pixels (LSP), List of Insignificant Sets (LIS), and List of Zero Trees (LZT), the method arranges these wavelet coefficients into a hierarchical tree structure. It then tests LSP coefficients for significance against a threshold and encodes their significance and sign. LIS coefficients with significant child coefficients are promoted to the LSP and zero trees are identified. The threshold is decreased by 0.5 for successive repetitions. Significance testing, encoding, and threshold modification are performed until the desired bit count or maximum distortion is reached. The compressed bitstream encodes the image's important information effectively, enabling large compression ratios without sacrificing visual quality.

Advantages of SPIHT:

• SPIHT's great compression performance aids in keeping the watermarked image's quality, which is a major plus.

• Adding the watermark to the wavelet coefficients prior to SPIHT compression makes the image more resistant to typical assaults in image processing, which improves its resilience.
• Invisibility: If the watermark is embedded correctly in the wavelet domain, it will not be visible to the naked eye.

3.4 SVD

As an orthogonal matrix decomposition technique, SVD is dependable and strong. The signal processing domain is seeing SVD's rising popularity for conceptual and stability-related reasons. When it comes to processing images, SVD is a desirable algebraic transform. In imaging, SVD's characteristics stand out. More research and contributions are needed for certain SVD features, even if they are completely used in image processing for others. Its capacity to approximate matrices of a given rank and its connection to the rank of a matrix are important properties of SVD. It is possible to characterise digital images by adding up a limited number of eigen images as they are often represented by low rank matrices. The idea of treating the signal as two separate subspaces gives birth to this notion [8]. The SVD/Rank based reduction was first found for square matrices in 1873 and 1874. In 1930, it is improved for rectangular matrices. The following equation represents the SVD of a rectangular matrix P.

$P=A \varepsilon B^T$                       (4)

Assuming that P is a rectangular matrix of size MXN.A and B are ortho-normal's rectangular matrices. The diagonal members of the diagonal matrix, denoted as ε, are singular values [19]. Since it is data dependent, SVD decomposition is unable to handle time-frequency domain data [21].

3.5 Watermarking

Watermarking is a technique that allows one image to be embedded with specific data into another image. When it comes to protecting sensitive information and intellectual property, it is the most important subfield of data security. On the other hand, there are two main varieties of image watermarking: invisible and visible. As a result, video images use the visible watermark, whilst audio and static signals make extensive use of the invisible watermark. Image watermarking, on the other hand, provides encrypted digital data transfer by enclosing the data inside the cover image [20]. To create a digital image with a watermark, the watermarking algorithm and a watermarking key are used in the embedding process. Different image domains, such as the space domain, the frequency domain, or the wavelets, call for different embedding methods. The suggested watermarking procedure begins with converting the watermark image to watermark bits; next, the watermark bits are mixed with the gold sequence; and finally, the watermark is embedded in the blocks at the chosen spot using this mixture.

File compression using QIM (Quantisation Index Module) helps make storage and transmission more economical by lowering the amount of bits required to represent data. Digital material may have watermarks embedded into it using QIM with a step parameter of 255. Because it enables the quantisation process to cover the whole range of grayscale values (0-255), a step size of 255 is often employed in QIM for grayscale since we are using images. This guarantees that the watermark may be included into the image with enough variance to be undetectable. This method makes sure the watermark is unreadable and resistant to manipulation and assaults. The overall architecture of the proposed work is depicted in Figure 2. The first step in the procedure is to create a binary watermark, which is a logo. The frequency components of the original image may then be separated by using SPIHT methods to translate it into the frequency domain. The modified image's coefficients are then quantized using two levels. It is computationally feasible to handle a 2-level decomposition, particularly for DICOM images, which often need to be processed quickly for medical purposes. Because DICOM images are useful for diagnosis, it is crucial to maintain detail in key areas. For medical imaging requirements, a 2-level decomposition delivers sufficient compression while preserving image integrity. It offers sturdy areas where a watermark may be inserted without significantly deteriorating the image. The quantization indices are changed in accordance with the watermark bits in order to embed the watermark: for a '1' bit, the index is shifted up, and for a '0' bit, it is moved down. The watermarked image is created by transforming the altered coefficients back into the spatial domain once the watermark has been embedded. The watermarked image undergoes the same transformation and quantization processes as before, and the quantization indices are examined to confirm the watermark's existence. This technique is a good fit for secure watermarking applications as it is resistant to many types of attacks and maintains a high level of visual clarity.

The following steps explain the proposed work.

• Open the medical image and set its parameters.
• Use a chaotic map (such as a logistic map) to create chaotic sequences.
• Utilizing the chaotic sequences, jumble and alter the pixel values to encrypt the medical image (watermark image).
• Set up the required settings for the FOA.
• Specify the goal function that will be used to optimize the encryption procedure.
• To improve encryption security, use FOA to optimize the chaotic map settings.
• For the final image encryption, apply chaotic settings that are optimal.
• Apply Discrete Wavelet Transform (DWT) for two level on the image that has been encrypted.
• Apply the DWT coefficients' SPIHT encoding again for 2 level as given in the architecture shown in Figure 2. SVD should be applied to the SPIHT-encoded image.
• Utilizing Quantization Index Modulation, include the watermark into the SVD components (QIM).
• Reassemble the watermarked image by using the DWT and inverse SPIHT transformations for extraction and decrypt to get the watermark image.

0402a.png

(a)

0402b.png

(b)

Figure 2. Architecture of (a) Encryption and watermarking process (b) Decryption and extraction process

Embedding process

The watermark may be embedded into the wavelet coefficients either before or during the SPIHT encoding process to integrate it into the SPIHT compression process.

• Apply a wavelet transform (such as DWT) to decompose the original image into multiple subbands (LL, LH, HL, HH). These subbands represent different frequency components of the image.
• Determine the amount of the chosen coefficients and tweak them to fit the watermark bit that has to be inserted. Several kinds of assaults may be thwarted by this strategy.
• Perform SPIHT encoding on the modified wavelet coefficients to compress the image.
• The SPIHT-encoded bitstream, now containing the embedded watermark, can be transmitted or stored efficiently.
• After SPIHT transformation the LL sub band is considered for SVD decomposition. While LH, HL, and HH include greater information, the LL sub band provides crude approximations. LL sub band embedding improves robustness towards compression.
• SVD applied to the encrypted watermark image and then the embedding process happens.

Extraction process

• Decode the SPIHT bitstream to reconstruct the wavelet coefficients of the image. Apply the inverse wavelet transform to obtain the watermarked image in the spatial domain.
• Extract the watermark from the wavelet coefficients using the same method and parameters used during embedding.
• Identifying variations in quantisation levels or coefficient magnitudes, is one embedding strategy that determines the precise extraction procedure.
• By integrating SPIHT with watermarking take use of effective image compression and implant strong, undetectable watermarks.

Image encryption and watermarking are critical in safeguarding digital content, especially with the rise of unauthorized access and distribution. Enhanced chaos-based encryption techniques have been widely adopted for their ability to create highly secure and unpredictable encryption keys. Chaos theory, which leverages the sensitivity to initial conditions, enables the generation of complex patterns that are nearly impossible to predict or replicate. In the context of image encryption, this enhanced chaos can be applied to pixel scrambling or key generation, resulting in encrypted images that resist various cryptographic attacks. The security provided by such methods makes them particularly suitable for high-stakes applications where data integrity and confidentiality are paramount. For tasks that need both safe copyright protection and high-quality image transmission, our hybrid solution shines.

Watermarking using SPIHT decomposition is a robust approach for embedding watermarks within an image. SPIHT is a wavelet-based compression technique that preserves the hierarchical structure of image data, allowing for efficient compression and progressive image transmission. By integrating watermarking within the SPIHT framework, the watermark can be embedded in the most significant wavelet coefficients, ensuring that it remains intact even after compression. This method enhances the imperceptibility of the watermark while maintaining the quality of the original image. Moreover, the hierarchical nature of SPIHT allows for multi-level watermarking, where different levels of wavelet decomposition can carry varying strengths of watermark signals, balancing robustness and transparency. Singular Value Decomposition (SVD) combined with Quantization Index Modulation (QIM) embedding further strengthens the watermarking process. SVD is a powerful mathematical tool used to decompose an image into its singular values, which can then be modified to embed a watermark. QIM is a robust watermark embedding technique that adjusts the quantization of the singular values to encode the watermark. The synergy between SVD and QIM ensures that the watermark is resilient to common attacks like noise addition, compression, and geometric transformations. Additionally, the SVD-based approach benefits from the stability of singular values, meaning that minor modifications do not significantly alter the visual quality of the image. This combined technique is highly effective in maintaining the balance between watermark imperceptibility and robustness, making it ideal for applications that require secure image transmission and authentication.