This paper introduces an adaptive approach for embedding digital watermarks into image, audio, and video files to protect multimedia content within modern information and communication systems. The proposed method leverages the host signal's spectral, statistical, and perceptual properties to ensure high perceptual fidelity and robustness against various attacks, including compression, filtering, and geometric distortions. A mathematical model for the watermarking process, based on a secret key, has been developed; this model formalizes the procedures for watermark embedding and extraction while ensuring minimal perceptual distortion. The system architecture is implemented as a multi-agent framework incorporating cryptographic protection, open Application Programming Interfaces (APIs), and automated integrity verification modules. Comprehensive testing conducted on real-world multimedia data has confirmed the effectiveness of this approach. The proposed solution is suitable for application in various domains, including digital forensics, copyright protection systems, streaming platforms, and digital archiving.
The rapid proliferation of digital multimedia content within information and communication systems has intensified the need for robust methods to protect such data from counterfeiting, unauthorized distribution, and loss of authenticity. This challenge is particularly acute for video and audio content transmitted over open communication channels or stored in cloud environments. To address these issues, digital watermarking (DWM) technologies have become a key solution, providing mechanisms for digital identification, authorship verification, data integrity control, and the tracking of content distribution sources.
Digital watermarks are imperceptible or minimally perceptible data structures embedded within a media object that convey auxiliary information—identifiers, hash values, or digital certificates — without degrading the user’s perceptual experience. In contrast to conventional watermarking schemes, which rely on static rules and predetermined parameters, adaptive methods dynamically configure the embedding algorithm based on a local analysis of a specific segment’s properties within the media file. This approach enables the dynamic adjustment of watermark parameters, such as embedding depth, spectral characteristics, spatial position, and intensity, in response to the signal’s energy spectrum, local image contrast, and models of human perception.
Therefore, adaptive digital watermarking algorithms provide high imperceptibility (imperceptible to human sight or hearing) and robustness to a wide range of attacks and transformations, including JPEG and MP3 compression, filtering, cropping, scaling, and reencoding, while ensuring minimal degradation of the original signal quality. Furthermore, there is a pressing need to counteract emerging forms of content manipulation, particularly deepfake technologies, neural reconstruction, and other generative models to create synthetic audio and video materials. Such fabricated content can be deployed for phishing, cyber blackmail, disinformation campaigns, or evidence tampering in digital forensics.
This study aims to develop and implement an adaptive method for embedding digital
watermarks designed to protect audio and video content within information and communication
systems. The proposed approach leverages spectral and perceptual analysis, in addition to
statistical and neural network models, to optimize embedding parameters according to the specific
characteristics of each media object (e.g., image, sound, video) [
This study also formalizes the architecture of an adaptive software framework that comprises modules for preprocessing, watermark embedding, detection, quality assessment, and integration with information and communication protocols. The architecture is founded upon the principles of modularity, scalability, and compatibility with common multimedia standards (e.g., H.264, AAC, JPEG2000), enabling the implementation of the proposed algorithms in on-premises environments and in cloud infrastructure or real-time streaming systems.
Consequently, the developed adaptive digital watermarking methods provide an effective solution for protecting multimedia content from unauthorized use, distortion, and loss of authenticity, while preserving high perceptual quality and ensuring suitability for integrating modern information and communication technologies. 2.
Modern research in digital watermarking increasingly incorporates adaptive and neural
networkbased methods to enhance robustness against attacks while preserving the quality of multimedia
content. For instance, Taha, Ngadiran, and Ehkan [
Quan et al. [
Li and Yue [
Chen et al. [
Naseem et al. [
This research confirms that the most effective multimedia information protection systems are based on integrating adaptive strategies, statistical signal analysis, and artificial intelligence. This synergy ensures high robustness, transparency, and authenticity for digital watermarks within the evolving digital information environment. 3.
This study integrates theoretical and applied methodologies, encompassing mathematical modeling, information security principles, and digital signal processing. The attack model incorporated a comprehensive set of transformations based on the discrete cosine transform (DCT) and the Fast Fourier Transform (FFT), geometric distortions such as scaling and rotation, StirMarktype benchmark attacks, and neural network-based reconstructions.
The adaptive embedding algorithm leverages the entropy characteristics, frequency-domain power, and texture features of the multimedia signal, which are extracted using techniques such as Gabor filters and wavelet analysis. Embedding regions are determined based on perceptual visibility models and weight functions derived from a trained convolutional neural network (CNN).
A Monte Carlo simulation with 1000 independent iterations was conducted to validate the method’s effectiveness. Each iteration involved the random selection of a multimedia container (in JPEG or WAV format), the generation of a unique embedding configuration, the determination of a random embedding location, a variable embedding density (from 1 to 6), and a specific attack scenario. The digital watermark was embedded and subsequently extracted in each iteration, and the results were evaluated.
Performance was assessed using a suite of quality metrics: Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index (SSIM), Quality Index (Q-index), Bit Error Rate (BER), Normalized Cross-Correlation (NCC), and Learned Perceptual Image Patch Similarity (LPIPS). A successful extraction was defined as an iteration in which the recovery accuracy, τ, met or exceeded a threshold of 0.98. 4.
Modern security tools and methods for ensuring the integrity of multimedia information are rapidly evolving in response to the escalating threats of unauthorized access, content modification, and copyright infringement within digital communication channels. A significant limitation of many existing solutions is their reliance on static architectures with fixed processing algorithms, which fail to account for the dynamic nature of the transmission environment or the specific content characteristics.
A comprehensive system for controlling the integrity of multimedia objects is typically
implemented as a software-hardware architecture comprising several key functional modules.
These include: an integrity verification module for authenticating digital signatures, calculating
hash functions, and verifying embedded watermarks [
Fig. 1 presents a contextual and structural Data Flow Diagram (DFD) model of an adaptive
digital watermarking system designed to protect image, audio, and video content within
information and communication systems. The architecture comprises five key functional
subsystems: a control interface and API, a content analysis module, an adaptive embedding module,
a verification module, and a secure repository for service data and logs [
To verify content integrity or authenticity, a user can initiate the verification process, during
which the system extracts the embedded watermark, compares it against the reference value, and
generates a corresponding response [
The application of adaptive digital watermarking methods, which leverage perceptual analysis
[
The escalating cybersecurity threats [19] to multimedia information necessitate a
comprehensive risk analysis of unauthorized access, distortion, forgery, and illicit audio, video, and
graphic content copying. This study presents a systematic classification of threats into primary
categories and proposes a feature-based model [
Fig. 2 presents a generalized scheme of classification features for multimedia threats, illustrating a hierarchical structure of risks that affect multimedia information security within information and communication systems. The threats are categorized into four primary groups: by context of occurrence (e.g., intentional attacks, unintentional distortions), by channel of influence (e.g., network interference, hardware failures), by attack vector (e.g., content modification, extraction of hidden information, insertion of malicious watermarks), and by enabling technology (e.g., compression artifacts, AI-based reconstruction, format conversion). This classification framework enables a systematic analysis of the risks associated with multimedia processing and transmission. It facilitates the development of adaptive digital watermarking strategies capable of effectively countering a wide range of attacks in the contemporary digital environment.
A formalized steganography model is defined as a set of two basic functions—embedding and extraction: 1. F : M ∙ B ∙ K → B ' is an embedding function that correlates a message m ∈ M , an empty container b ∈ B, and a secret key k ∈ K with a modified container b ' ∈ B ' . 2. F ' : B ' ∙ K → M is an extraction function that, given the modified container b ' and key k recovers the original message m.
Thus, an adaptive steganographic system is defined as the five F , F ' , M , B , K , where M is a set of messages, B is a set of containers, K is a set of keys, F and F ' are embedding and extraction algorithms, respectively. This allows us to formally describe the stage for generating a protected media object and the procedure for verifying integrity or authenticity. This approach ensures generalization of processes at the model level, which is necessary for building universal software solutions. In addition, the system structure allows you to set adaptation parameters depending on the type of content and threats. The formalization based on sets and functions also facilitates integration with cryptographic protocols and access control mechanisms.
A comparative analysis of patented and published solutions in digital watermarking has revealed several systemic limitations of traditional methods that restrict their practical application. Key disadvantages include the introduction of significant visual or acoustic distortions, the presence of detectable embedding artifacts vulnerable to steganalysis, the potential for extracting hidden data without key knowledge if the modified container is compromised, and a lack of robustness against watermark substitution or distortion resulting from minor edits to the multimedia file.
In contrast, modern adaptive embedding methods which leverage perceptual modeling [
Effective digital multimedia watermarking systems must account for a wide range of potential attacks targeting both the steganographic methods and the embedded watermarks. Threats from passive (e.g., observation, steganalysis) and active (e.g., modification, deletion, substitution) adversaries operating within a dynamic digital environment pose significant challenges.
Fig. 3 illustrates the structural architecture of an adaptive digital watermark embedding system designed to protect image, audio, and video content within information and communication systems. The diagram outlines the primary stages of the watermarking lifecycle: message encryption using the Advanced Encryption Standard (AES) symmetric block cipher; integrity tag generation via a Hash-based Message Authentication Code (HMAC); adaptive embedding into the multimedia container; transmission over a channel susceptible to potential threats; and subsequent extraction, verification, and decryption on the recipient’s side. The architecture also accommodates the integration of auxiliary components, including a Digital Rights Management (DRM) module, a Security Information and Event Management (SIEM) system, and an event logging subsystem. All data flows within the system are labeled with unique identifiers (e.g., Data1-Data11, Log1-Log4, Auth1) to provide clear visibility into the interaction logic and enhance the transparency of information process control. This comprehensive architecture ensures robust protection, scalability, and ease of integration into modern digital environments.
Assessing the resilience of such systems necessitates a comprehensive approach. This study
employs attack models based on extended threat matrices, encompassing various types of
interference, modification intensities, and access to keys and data [
As a result of this research, mathematical models and adaptive methods for monitoring the
integrity of multimedia information, based on the covert embedding and extraction of digital
watermarks, have been developed [
A key feature of the model is the concept of hiding space, a linearized sequence of elements of a
multimedia container suitable for marking, determined based on the signal’s type, format, and
perceptual significance [
P ij = P ( s iH −s iL ) , n P ( l i )= i , l i ∈ { 1,2 , … , 2k } .
∑ n P ( k i )= mi ,
∑ m where each parameter belongs to one of three groups.
Statistical analysis of paired elements of the hiding space—evaluates the distribution of differences between the high and low bits of the elements: where ( s iH −s iL ) are the high and low bits of the elements, respectively.
Frequency analysis of bit series—determines the probability of occurrence of a bit series of length l in the hiding space: (1) (2) (3) (4)
Analysis of the lengths of identical bit sequences—the probability of a series with identical bits: where mi is the number of series of length i .
Based on this model, adaptive methods for embedding and extracting digital watermarks have been developed that depend on the secret key k ∈ K , which uniquely determines how the mark is placed in the hiding space. The embedded bits are distributed according to the patterns generated from the key sequence, ensuring uniformity and attack resistance. This approach makes it impossible to reproduce or modify the mark without knowing the key and minimizes the likelihood of an attacker localizing the marking structure. In addition, the dependence of the embedding positions on the key ensures cryptographic reliability and makes it easy to scale the system to fit different container sizes. In the event of an attempt to remove or edit the watermark, the system reacts with a loss of correctness during authentication, which allows you to identify the fact of interference quickly.
The function describes the direct adaptive steganographic transformation:
F : M ∙ B ∙ K → B ' , b'i = { mi , if k i−1=sh bi , otherwise , (5) where F is an adaptive embedding function that may depend on spectral (DCT), wave (DWT), or spatial (LSB) criteria, M is a set of messages to be embedded, B is a set of containers, K is a set of keys, bi is the element of the original container’s hiding space, b'i are elements of the modified container, mi are bits of the embedded message, k i are blocks of the key sequence generated from the secret key, s h mask template that defines the places for embedding. Thus, the adaptive steganographic system dynamically generates the space for embedding based on statistical and structural features of the signal, while ensuring a high level of transparency, integrity, and cryptographic strength when transmitting multimedia data in open networks.
Fig. 4 shows a spatial vertical model of the embedding F and extraction F′, functions that implement adaptive digital watermarking in multimedia containers. Visually, the diagram is divided into three zones: input data (left), processing logic (center), and results (right). In the process of embedding, a message m is hidden in the elements of a container bi using a key sequence k i according to a template rule. The extraction of information follows a similar pattern, with the message being recovered from the modified container b'i. Data and key flows are defined and labeled for all stages, which ensures a transparent formalization of the process and increases the system’s reproducibility.
Integrity control is realized through adaptive embedding of digital watermarks, which allows
you to verify the authenticity of a multimedia container without the need to store separate control
hashes or reference copies. The method uses cryptographic hash functions that provide data
verification in the presence of only a modified container and a secret key [
Solving the problem of building a model for assessing the distortions introduced during the
embedding of a digital watermark and methods for their quantitative analysis is the content of a
separate research stage. The quality of a multimedia container is assessed using the distance metric
between the original (unmodified) container I and the modified (embedded) container O , which is
formalized as [
q = K ( I , O ) , where K is the generalized similarity metric function (e.g., SSIM, PSNR, VMAF, NCC, etc.).
The proposed quality model is presented as a set of indicators:
S = { f i ( I , O ) ∨ i =1 , … , n } , (7) where n is the model dimension, f i is the corresponding quality metric. For a specific type and format of a multimedia container (image, audio, video), the optimal subset of relevant criteria is selected:
Θ = { f i ( I , O ) ∨ I , O ∈ B } , where B is a set of valid multimedia data formats.
To ensure cryptographic security, a key function is used G [
K → k = { k 1 , k 2 , … , k n } , k i = HMA C K ( i ||t ||UID ) , i ∈ N , t =timestamp , UID =user_id}.
To guarantee confidentiality, the message m is encrypted before embedding using AES: c = AES k ( m ) , c → is embedded instead of m .
The pull function F ' : B ' ∙ K → M is defined as:
mi = { F ( b'i ) ' , if k i−1=sh ⊥ , otherwise , where ⊥ is a marker indicating the absence of an embedded bit in this element.
Verification is performed using the integrity function:
Valid = { true ' , if HMAC K ( c )=stored_tag false, otherwis
A composition of functions can describe the complete adaptive labeling process: where Enc is a function of message encryption before embedding.
After extracting the recovered message: where Dec is an AES decryption.
Reverse quality and compliance checks can be performed:
W ( M , B , K , sh )= { b'1 , … , b'n } , b'i = F ( Enc ( mi ) , bi , k i ) , mi = Dec ( F ' ( b'i ) ) ,
n Match ( m , ^m)= ∑ δ ( mi =^mi ) /n ≥ τ , i=1 (8) (9) (10) (11) (12) (13) (14) (15) (16) (17) where τ is a threshold value of recovery accuracy (for example, 0.99).
The proposed formalized approach provides a transparent mathematical model for operating an
adaptive digital watermarking system. It demonstrates its practical suitability for deployment in
dynamic information and communication systems [
An architecture for an adaptive software framework for maintaining the integrity of multimedia
information has been developed, founded upon a digital watermarking model [
Each agent is constructed using a modular architecture, comprising a main software module and
a set of functional components with a clear distinction between mandatory modules (required for
basic operation) and optional modules (which can be connected based on media type, security
policy, or user requirements) [
A software framework has been developed to ensure the integrity of multimedia information
using adaptive digital watermarking, with a focus on modern graphic (JPEG, PNG), audio (WAV,
MP3), and video (MP4, MKV) formats [
Structurally, the framework is composed of four primary agents. A management and
monitoring agent coordinates inter-component interaction, configures security settings, and
maintains event logs supporting SIEM/SOAR systems [
A suite delivers the system’s functionality through specialized modules. A cryptographic
protection module (crypto_module) implements adaptive encryption based on AES-256 and
HMAC-SHA256, while a watermark insertion/extraction module (adaptive_stego) provides
adaptive embedding using DCT, FFT, and LSB methods [24]. A container naturalness evaluation
module (perceptual_model) analyzes perceptual quality using SSIM, PSNR, and the modern LPIPS
metric, while a distortion evaluation module (distortion_eval) determines acceptable modification
levels that do not compromise quality [
To evaluate the effectiveness of the developed method, a test sample was formed, comprising
1,000 JPEG images and 1,000 WAV audio files [
The results of the experimental study confirm that the proposed method demonstrates high
efficacy with minimal impact on the perceptual quality of the multimedia content. This conclusion
is supported by stable values for the Structural Similarity Index Measure (SSIM) and Perceptual
Evaluation of Speech Quality (PESQ) metrics [20]. The adaptive approach facilitates an optimal
balance between watermark robustness against modifications and preserving the perceptual quality
of images, audio, and video [
Fig. 6 presents a Data Flow Diagram (DFD) of the adaptive multimedia integrity control
software framework, which is architected around the interaction of four specialized agents: a
control and monitoring agent, a storage agent, a multimedia registration agent, and an integrity
verification agent. The user initiates the configuration process via the control agent, which
orchestrates the interaction among all other system components [
The implemented system validates the effectiveness of modern approaches to the embedding and controlling digital watermarks, confirming its suitability for integration into information and communication systems to ensure the digital authenticity, integrity, and protection of multimedia objects.
Within a simulation modeling framework implemented using the Monte Carlo method, a comparative analysis was conducted to evaluate the effectiveness of the proposed adaptive watermarking method against standard approaches, specifically those based on the least significant bit (LSB) and static insertion using the discrete cosine transform [17, 19, 21, 26, 27]. The Monte Carlo method was employed to simulate many random scenarios for embedding watermarks into multimedia containers, which facilitated an assessment of the methods’ stability and quality under variable input conditions. The container (JPEG or WAV), embedding density (from 1 to 6 bits per block), watermark position, and attack type were selected randomly in each simulation. This approach provides a statistically robust evaluation of effectiveness, enabling the determination of average values for key metrics such as SSIM, PESQ, and BER, as well as their stability under realworld conditions.
Based on the simulation results, a heatmap was constructed to illustrate the dependence of the Structural Similarity Index Measure (SSIM) on both the multimedia container format (JPEG or WAV) and the digital watermark embedding density [22]. This visualization facilitates an evaluation of the impact of these embedding parameters on perceptual content quality across a wide range of conditions. The results indicate that at embedding densities up to 4 bits per block, the proposed adaptive method maintains high imperceptibility (SSIM > 0.98) and a low bit error rate (BER < 0.02), significantly outperforming classical LSB and static DCT-based watermarking methods [25–27]. Furthermore, the average processing time remains within limits that are suitable for practical implementation in real-time information and communication systems.
Fig. 8 presents a heatmap illustrating the relationship between the Structural Similarity Index
Measure (SSIM), the multimedia container type (JPEG images and WAV audio), and the digital
watermark embedding density (ranging from 1 to 6 bits per element). The color scale represents the
degree of structural integrity preservation, where SSIM values approaching 1.000 signify nearly
imperceptible distortions introduced by the watermarking process. The results indicate that the
adaptive method maintains high perceptual quality (SSIM > 0.98) at embedding densities up to 4
bits per element for both image and audio containers. A further increase in density to the 5–6 bit
range results in a slight decrease in SSIM values, indicating the emergence of noticeable perceptual
artifacts [
Fig. 9 presents a graph illustrating the dependence of the Structural Similarity Index Measure (SSIM) on the digital watermark embedding density for two multimedia container formats: JPEG (image) and WAV (audio). As the embedding density increases from 1 to 6 bits per element, a gradual decrease in SSIM values is observed, corresponding to an increase in perceptual distortions.
Despite a high embedding density of up to 6 bits per unit of information, testing demonstrates that the SSIM values—an indicator of structural similarity between the original and watermarked signals—remain high, exceeding a threshold of 0.95. This indicates that even with intensive data embedding, the adaptive algorithm maintains high perceptual fidelity without compromising the visual or acoustic quality of the content. This level of transparency is critical for applications in information and communication systems, where multimedia data quality is crucial for the end user.
Particular attention is given to a comparative analysis of watermarking effectiveness in different container types, specifically graphic (JPEG) and audio (WAV) formats. The results show that WAV audio containers provide greater robustness for embedded watermarks, particularly against attacks such as re-encoding, filtering, and noise addition. This is attributable to the higher internal redundancy and lossless nature of WAV files, which affords greater flexibility for embedding data without a significant risk of corruption. Furthermore, the spectral characteristics of audio signals in the WAV format are more stable and predictable than the high-frequency components of JPEG images, which are highly susceptible to quantization during compression, allowing the watermark to remain intact even after subsequent signal transformations.
On the other hand, despite their ubiquity, JPEG graphic containers possess less redundancy, especially in the high-frequency regions of an image that are often aggressively quantized or distorted during compression. This creates significant challenges for preserving the embedded watermark, as some data bits may be destroyed or altered. In this context, an adaptive watermarking approach that analyzes the local properties of the container—such as texture complexity, luminance, and contrast—enables the selection of optimal embedding regions that are most resilient to alterations from digital processing.
The built-in mechanism for the dynamic adjustment of parameters, including embedding depth, masking coefficients, and spatio-spectral localization, allows the adaptive technique to demonstrate high flexibility and stability in both the visual and acoustic domains. This not only preserves the integrity of the watermark but also ensures that the content meets the requirements of perceptual transparency and the absence of perceptible visual or auditory artifacts.
The graph in Fig. 10 compares digital watermark stability in JPEG, WAV, and MP3 formats when using classical versus adaptive methods. The results show that adaptive approaches provide superior watermark preservation, particularly in the WAV format, which is attributed to the container’s internal redundancy and stable spectral properties.
The generalized results confirm the adaptive digital watermarking algorithm’s superiority in protecting multimedia content within modern information and communication systems. Its capacity to maintain high SSIM values even under aggressive embedding conditions, combined with its enhanced stability in audio formats—particularly WAV—compared to JPEG graphics, underscores the viability of this approach for practical applications in digital identification, authorization, and media protection in open networks.
Despite an intensive embedding density of up to 6 bits per data unit, experimental results confirm that the media content’s high visual and acoustic quality is preserved. In particular, the Structural Similarity Index Measure (SSIM), a metric used to assess the similarity between original and modified image or video frames, remains consistently high, exceeding 0.95. This indicates that the proposed adaptive method does not introduce significant distortions into the signal’s structure, thereby preserving its naturalness and rendering the changes imperceptible to the user. This is a critical feature for practical applications in information and communication systems where perceptual quality is a priority.
An additional comparative analysis of watermark robustness across different media container types revealed that WAV audio files provide a more resilient environment for watermark integration than JPEG graphic images. This can be attributed to several key factors. First, WAV containers possess a significantly higher degree of internal redundancy, which provides lowentropy regions suitable for inconspicuous embedding without disrupting the overall signal structure. Second, the spectral characteristics of audio signals in the WAV format are more stable and less sensitive to local alterations, allowing the embedded information to be preserved even after subsequent processing or playback.
In contrast, JPEG containers, despite their prevalence, are more vulnerable to watermark information loss due to the inherent nature of lossy compression algorithms. This is particularly true for high-frequency components, which are often aggressively quantized during encoding, making it difficult to maintain a stable watermark. In this context, adaptive methods demonstrate their advantages through their ability to dynamically select optimal embedding regions by analyzing local signal properties such as energy, texture features, or spectral density. This allows for an optimal balance between robustness, imperceptibility, and embedding capacity.
Thus, adaptive methods of embedding digital watermarks demonstrate high efficacy for both image and audio files, ensuring robustness against distortion, maintaining high SSIM values, and proving their suitability for practical implementation in a wide array of digital content protection systems, including multimedia platforms, streaming services, authorization systems, and digital forensics.
This paper presents a comprehensive, adaptive method for embedding digital watermarks to protect image, audio, and video files within information and communication systems. The proposed approach is founded upon leveraging the multimedia signal’s statistical, spectral, and perceptual characteristics, ensuring the embedding process’s dynamic adaptation according to the content type and specific processing conditions.
The developed mathematical model formally describes the watermark embedding and extraction processes, ensuring their robustness against primary attack vectors, including compression, geometric distortions, and neural network-based attacks. The proposed software architecture is implemented as a scalable, multi-agent system featuring support for logging mechanisms, cryptographic protection (AES, HMAC), SIEM/DRM infrastructure integration, and artificial intelligence modules for assessing perceptual quality.
The results of practical testing on a large sample of JPEG and WAV files demonstrate the high efficacy of the approach. At an embedding density of up to 4 bits per block, the SSIM value exceeds 0.98, indicating that high perceptual quality is maintained while ensuring watermark integrity. Importantly, the authentication procedure is blind, requiring only the watermarked container and the secret key, thus obviating the need to store a reference copy.
The proposed solution holds significant practical value for applications in digital forensics, copyright protection, video streaming, and distributed digital authentication systems. Future work will incorporate blockchain technologies to create an immutable audit trail of watermarking events, extend support to modern video streaming formats such as HLS and MPEG-DASH, and integrate Explainable AI (XAI) models for integrity verification within critical information infrastructure environments.
Despite the positive results, this study has several limitations. Specifically, the method’s robustness against re-encoding into modern video formats like H.265 and AV1 and its efficacy in the presence of active noise associated with transcoding or hardware recording artifacts were not evaluated. Future research will be directed at overcoming these limitations and developing intelligent, adaptive schemes for the real-time verification of digital watermarks. Declaration on Generative AI While preparing this work, the authors used the AI programs Grammarly Pro to correct text grammar and Strike Plagiarism to search for possible plagiarism. After using this tool, the authors reviewed and edited the content as needed and took full responsibility for the publication’s content. [17] P. Kadian, S. M. Arora, N. Arora, Robust digital watermarking techniques for copyright protection of digital data: A survey, Wireless Pers. Commun. 118(4) (2021) 3225–3249. doi:10.1007/s11277-021-08177-w [18] P. Aberna, L. Agilandeeswari, Digital image and video watermarking: Methodologies, attacks, applications, and future directions, Multimed. Tools Appl. 83(2) (2023) 5531–5591. doi:10.1007/s11042-023-15806-y [19] A. Saini, S. Bhardwaj, A review on digital video watermarking security: Significance and persistent challenges, in: Int. Conf. on Trends in Quantum Computing and Emerging Business Technologies, Pune, India, 2024, 1–8, doi:10.1109/TQCEBT59414.2024.10545079 [20] Z. Wang, et al., Deep image steganography using transformer and recursive permutation,
Entropy, 24, 2022, 878. doi:10.3390/e24070878 [21] R. R. Sunesh Kishore, A. Saini, Optimized image watermarking with artificial neural networks and histogram shape, J. Inf. Optim. Sci. 41(7) (2020) 1597–1613. doi:10.1080/02522667. 2020.1802131 [22] Y. Kostiuk, et al., Information and intelligent forecasting systems based on the methods of neural network theory, in: IEEE Int. Conf. on Smart Information Systems and Technologies (SIST), 2023, 168–173. doi:10.1109/SIST58284.2023.10223499 [23] B. Jagadeesh, D. Praveen Kumar, Fuzzy-neuro based robust digital image watermarking technique, Int. J. Adv. Res. Comput. Commun. Eng. 3(7) (2014) 7380–7385. [24] A. Kumar, T. V. Narayana Rao, Digital image watermarking using fuzzy logic and genetic algorithm, Int. J. Comput. Trends Technol. 41(2) (2016) 101–105. [25] S. J. Horng, et al., An adaptive watermarking scheme for e-government document images,
Multimed. Tools Appl. 72 (2014) 3085–3103. doi:10.1007/s11042-013-1579-5 [26] A. Attaullah Javeed, et al., Watermarking technique for copyright protection of digital images using coupled differential equations, Multimed. Tools Appl. 84 (2025) 11027–11039. doi:10.1007/s11042-024-19337-y [27] P. Bhinder, N. Jindal, K. Singh, An improved robust image-adaptive watermarking with two watermarks using statistical decoder, Multimed. Tools Appl. 79 (2020) 183–217. doi:10.1007/s11042-019-07941-2