The proliferation of easy-to-use image editing software has led to a surge in manipulated images, particularly copy-move forgeries (CMF), where a part of an image is duplicated and pasted within the same image to mislead viewers. This poses a significant challenge for digital image authentication, critical in the legal, journalistic, and scientific domains where image integrity is paramount. While Convolutional Neural Networks (CNNs) have significantly improved copy-move forgery detection (CMFD), many existing approaches continue to face difficulties when dealing with intricate manipulations and varied dataset characteristics. In addition, the potential benefits of using dataset heterogeneity for improved detection have not been thoroughly examined. To bridge these gaps, this paper introduces Satyachitra, a CNN-based architecture specifically designed to identify subtle forgeries in complex image scenarios. The approach integrates pre-processing stages, including feature extraction, image segmentation, and forgery localization, to achieve robust detection. By analyzing data sets such as MICC-F220, MICC-F600, and a combined variant, we gain deep insights into the features of the data sets and their impact on the detection performance. The proposed system compares original and forged images using the CNN framework, supported by experiments with differential training-validation sets and randomly labeled data to assess model resilience. Comprehensive comparisons with existing models showcase the superiority of the proposed approach, achieving a detection accuracy of 90% within a predefined iteration limit, thus establishing its reliability and effectiveness.

In secure communication, simply encrypting a message is no longer sufficient. Modern systems must protect the content, verify the sender’s identity, and conceal the very presence of sensitive information. However, most existing hybrid cryptography–steganography approaches either rely heavily on shared symmetric keys or primarily address confidentiality, leaving vulnerabilities in authentication, key management, and detectability. To overcome these limitations, this paper presents a hybrid framework that combines public-key cryptography with a key-driven LSB steganographic process. The proposed system is designed to unify confidentiality, sender authentication, and covert transmission within a single workflow, while also improving resistance to steganalysis through randomized embedding. The method ensures that only a legitimate receiver can retrieve the hidden ciphertext, and only the genuine sender can be verified during message recovery. This layered design minimizes the risks associated with key exchange and predictable embedding patterns. Experimental evaluation using standard test images demonstrates a significant improvement in imperceptibility, achieving a PSNR of over 73 dB, near-perfect SSIM, and almost identical entropy and histogram characteristics compared to the original images. These outcomes confirm that the embedded data remains visually and statistically undetectable. The results also show reliable message extraction and stable security behavior, indicating that the proposed technique not only strengthens authentication but also achieves a high level of covert communication and robustness compared to conventional hybrid approaches.

Distributed Denial of Service (DDoS) attacks have gained popularity among cybercriminals as a favoured method of disruption. Application layer DDoS attacks are particularly intricate, as they overload web servers with re-quests, rendering them inaccessible to legitimate users and causing availability issues. These attacks are challenging to detect through network and transport-level security measures, making them even more concerning. This paper explores various categories of DDoS attacks, encompassing volumetric and protocol-focused attacks, with a particular focus on application-layer attacks, classifying them into Protocol-specific attacks and Generic attacks. It also delves into diverse defence strategies tailored to combat related attacks, such as HTTP Flood, DHCP starvation, SlowLoris, and others. Unlike earlier surveys, which centre on vulnerability-oriented taxonomies through 2017–2020, this work introduces an explicit, criterion-based comparative evaluation framework for attacks and defences, and extends the taxonomy with post-2020 developments containerized and cloud native low-rate attack surfaces, machine learning-driven detection and adversarial evasion, and zero-trust-based mitigation illustrated with the 2023 HTTP/2 ‘Rapid Reset’ incident. Finding that detection-only mechanisms still dominate current defences, the paper identifies recurring bottlenecks and proposes con-crete future-research directions, including detection resistant to adversarial machine learning and low-rate attack detection in containerized and serverless environments.

Face liveness detection plays an essential role in modern biometric authentication systems, especially as these systems are increasingly targeted by presentation attacks such as printed photographs, replayed videos, and high-quality screen displays. To address this challenge, this paper presents a lightweight and deployment-friendly face liveness detection approach that combines a compact convolutional neural network (CNN) with depth-based analysis. The proposed model operates on standard RGB images and requires only minimal preprocessing, making it suitable for real-time and latency-sensitive applications on consumer devices.The CNN is designed to effectively capture discriminative facial features that help separate real faces from spoofed ones, while monocular depth estimation is used to identify flat or non-3D facial artefacts commonly observed in spoof attacks. The system produces probabilistic outputs, enabling flexible threshold selection based on security requirements. Experimental evaluation is conducted on a dataset of 3866 samples, where the proposed approach achieves an overall accuracy of 93.59% and a ROC-AUC score of 0.9861. In addition, the model demonstrates strong and balanced per-class precision and recall for both real and fake classes. These results indicate that the proposed method offers a practical balance between accuracy, interpretability, and computational efficiency, making it suitable for real-world face authentication scenarios.

Steganography plays an important role in transferring secret data over insecure networks. Moreover, digital images are taken as a cover to communicate sensitive data. This paper works with the novel idea of a data-hiding technique in cover video files leveraging a combination of a Linear Feedback Shift Register (LFSR) and the Hénon 2D chaotic map. We utilized the chaotic theory and LFSR concepts to embed data within individual frames of the cover video, enhancing security by randomising the frames used and subsequent choice of pixels in chosen frames. Following that, the proposed algorithm makes use of LSB encoding to embed secret data in chosen pixels. The utilization of chaotic theory, rooted in nonlinear dynamics of physics, underscores the sensitivity of chaotic systems to minute changes in initial conditions, rendering it exceedingly challenging to discern the data’s embedding position without knowledge of the chaotic map’s initial value, ensuring high level in randomness in data embedding positions. Peak Signal Noise Ratio (PSNR), Mean Square Error (MSE) and Correlation measures are used for quantitative analysis of the proposed scheme, along with a comparison of existing methodologies to conclude that the proposed video steganography approach is novel and provides an efficient level of security.

Detecting Static malware is still a Critical Research challenge as a result of increase in obfuscation, polymorphism, and counteractive methods of evading. We present in this paper an optimized static-analysis pipeline which is based on the EMBER 2018 dataset and an improved ensemble of famous gradient boosted models named as CatBoost, LightGBM, and XGBoost. This system eliminates the need for heavy preprocessing and makes use of boosted decision trees, weighted Soft voting, optimized threshold strategy focusing to maximize recall and balancing reliable precision. The threshold search ($0.7^{*}$ Recall+0.3*Precision) significantly improves the model's ability to detect malicious files. Doing some experiment on the 600,000 training samples and 200,000 testing samples demonstrates effective balance between recall and precision, which can be supported by graphical analysis. The results confirm that carefully adjusted ensembles can do better than traditional approaches and can also be used for deployment in the real world.

Raw accuracy fails to reflect model performance in intrusion detection systems under concept drift and severe class imbalance. This discrepancy is evident on the UNSW-NB15 dataset where a majority-class classifier that never detects normal traffic achieves 95.50% accuracy compared with 97.64% accuracy achieved by a Hoeffding Tree with a 2.14 percentage point gap. Under balanced accuracy, the same comparison yields a 39.77 percentage point gap. A streaming evaluation is conducted on the UNSW-NB15 dataset with controlled synthetic drift, using a phase-based framework that separates pre-drift, drift, and post-drift behavior across 36 rolling windows. Across this setting, Adaptive Random Forest and Hoeffding Tree recover within one window after drift, while the static Random Forest does not return to its pre-drift performance. Gaussian Naive Bayes reaches a false positive rate of 1.0 in multiple windows during the drift phase. These results show that balanced accuracy, per-class recall, and false positive rate expose failure modes that raw accuracy does not capture under non-stationary, imbalanced conditions.

The exponential growth of patient medical data poses critical challenges in privacy, secure access, and efficient sharing. Existing centralized healthcare systems are prone to privacy leakage, data loss, and weak access control, often compromising patient anonymity. To address these issues, this article presents a blockchain-based healthcare framework that provides patient-centric control through a consent-driven access mechanism, identity separation, and decentralized storage. Personal identifiers are decoupled from medical records, while smart contracts enforce fine-grained and verifiable access policies. To preserve anonymity and enhance resilience, identifiers are obfuscated and single points of failure are eliminated. Immutable audit trails further strengthen accountability and mitigate the vulnerabilities inherent in centralized systems. Experimental evaluation shows that the proposed system achieves a communication cost of 1568 bits and a storage cost of 953 bits, both significantly lower than the existing schemes. These results demonstrate the efficiency, scalability, and fault tolerance of the model. By combining transparency, security, and lightweight overhead, this patient-centric approach establishes a practical and secure framework for managing sensitive medical data in the evolving digital healthcare landscape.

The combination of digital methods and biological principles has created new avenues for safe communication in the rapidly changing field of data security. This study introduces a novel steganographic technique that embeds plaintext messages in digital image pixel points by encoding them into DNA sequences. Using predetermined encoding rules, the suggested method first converts textual data into a binary format and then into synthetic DNA sequences. To reduce perceptual distortion, these DNAencoded messages are then hidden inside the least significant bits (LSBs) of specific pixels image. Combining conventional image steganography with DNA-based encoding improves robustness and data capacity.

In this constantly evolving field of cybersecurity, protection of sensitive information and computer systems remains the most pressing challenge. Achieving it by striking a balance between seamless access for authorized users and strong defenses against unauthorized intrusions. While traditional methods of authentication, which rely on usernames and passwords, have been dominant in this field for years, they come with significant vulnerabilities such as brute force attacks, phishing, and password theft. In response, methods of biometric authentication have become prevalent with the emergence of keystroke dynamics as a promising alternative. This study addresses the development of keystroke dynamics-based authentication methods that are enhanced using machine learning algorithms. Experimental results point to the efficiency of these models, with the CTGAN model reaching 99.9

End-to-end encryption (E2E) has become standard for secure messaging, but its protection does not cover the moments when a message is being written or read. This leaves a significant security gap on devices where messages are composed and viewed. At these endpoints, input tools like browser-based forms or web mail clients handle all the text that is typed, and they are usually trusted without a second thought. However, these input methods can be weak points due to unclear data collection practices, security flaws, or malicious extensions. This means a user's private message, password, or financial detail is often unprotected long enough for malware, malicious scripts, or other web threats to spy on it. In this research, a generalized method for pre-send encryption is introduced, demonstrated through a proof-of-concept called KryptBoard. The core idea is a user-controlled “Encrypt Mode” that can be implemented across different platforms. It focuses on a web browser extension implementation, which uses the ChaCha20-Poly1305 encryption scheme to ensure messages are both private and authentic. The resulting ciphertext can be placed into any web input field, making it compatible with any web application and adding a “double-lock” when used with existing E2E protocols. By addressing a long-overlooked security weakness, it is shown that strong, user-controlled endpoint protection can be achieved without compromising usability

With the rapid expansion of digital communication systems, ensuring the confidentiality of sensitive information has become a critical challenge. Image steganography offers an effective solution by concealing secret data within digital images such that the existence of the message remains hidden. This work proposes a secure multi-image Least Significant Bit (LSB) steganography framework enhanced with lightweight XOR-based encryption. In the proposed method, a long secret message is divided into multiple segments, encrypted using a randomly generated XOR key, and embedded into the least significant bits of different cover images. Distributing encrypted data across several images significantly improves security, as compromising a single image does not reveal meaningful information. Moreover, this approach enables high payload capacity while maintaining excellent visual quality. The performance of the proposed system is evaluated using Peak Signal-to-Noise Ratio (PSNR), entropy analysis, and histogram comparison. Experimental results demonstrate that the stego images remain visually indistinguishable from the original images, ensure accurate message recovery, and show strong resistance to statistical steganalysis. These characteristics make the proposed technique suitable for secure communication, data protection, and digital watermarking applications.

The rapid transition toward online exchange of information and digital storage has increased the need for strong protection of users information in healthcare sectors for storing patient records. As hospitals and clinics rely more on interconnected systems to manage data, the risk of unauthorized access, accidental exposure, and deliberate misuse also becomes higher. To address these concerns, this paper introduces HealthGuard Secure, a practical system that applies a chaotic map-based technique for protecting medical images and related documents. The system uses password-based authentication along with a fully reversible encryption process, allowing only verified users to retrieve the original files. During encryption, each medical image is converted into an unintelligible form that cannot be reconstructed without the correct parameters. The method preserves the accuracy of the original medical data and does not depend on specialized hardware, making it suitable for everyday use in healthcare environments. The application also provides a straightforward interface that supports easy uploading, encrypting, storing, and decrypting of files. Experimental results confirm that the approach offers strong security, low computational cost, and consistent performance across different types of medical images, making it a reliable option for safeguarding sensitive healthcare information.

Abstract Steganography involves the creation of stegosystems that conceal data within a cover medium. In this work, we propose a linguistic stegosystem that employs natural language as the concealment medium using a Word2Vec-based continuous bag-of-words (CBOW) model with noise contrastive estimation. The system selects from the top 8 predicted candidate words at each embeddable position, achieving an embedding capacity of up to 3 bits per word (bpw) and a practical embedding utilization of 44.83% (up to 50% maximum capacity) of eligible cover words. Experimental evaluation demonstrates a perplexity of 10.3, outperforming several existing linguistic steganography approaches while maintaining high semantic coherence. Steganalysis experiments show limited detectability, with detection accuracies of 58% using TS-RNN, 64% using a BERT-based classifier (AUC = 0.70), and a noise-based perturbation flag rate of 12%, indicating strong resistance against statistical and neural detection methods. The proposed approach balances embedding capacity, linguistic naturalness, and security while maintaining computational efficiency through the lightweight CBOW architecture.

In the digital era, where duplication and manipulation of multimedia content are becoming a lot easier, ensuring authenticity and confidentiality has become essential. This research introduces a lightweight hybrid technique that combines visible watermarking with encrypted steganography to enhance media security. A visible logo is embedded into the top-right corner of a cover image using simple pixel replacement, while an encrypted secret message is hidden within the green channel of the logo using randomized Least Significant Bit (LSB) substitution. The logo is proportionally resized and embedded without alpha blending for efficiency. Secret messages are converted to binary, encrypted, and inserted at pseudo-random positions in the green channel using a fixed seed to ensure reproducibility. The final watermarked image is evaluated using PSNR, SSIM, and Shannon entropy, all of which indicate high visual fidelity and secure message concealment. This method offers a balanced solution between perceptual quality and data security, making it well-suited for applications that require lightweight watermarking integrated with confidential data hiding.

Ensuring safe and secure transmission of sensitive information is becoming increasingly difficult due to expansion of networks and evolving threats. Older cryptographic techniques like AES offer confidentiality; however, they make encrypted data detectable, whereas steganography introduces a hidden layer but often falls short in payload capacity and vulnerability. This work proposes a new framework that combines authenticated encryption with an adaptive transform-domain embedding approach to overcome existing limitations. The process involves encrypting different text categories using AES-256 in Galois/Counter Mode(GCM) for confidentiality and integrity, followed by data embedding in a processed cover image using a hybrid DCT transform. An edge-based masking mechanism targets complex image regions to minimize distortion. Experimental evaluation shows that this combination of authenticated encryption and adaptive transform domain embedding achieves high imperceptibility as PSNR value is comparatively higher than previous versions with near-perfect SSIM, and strong histogram similarity. Also, the entropy of the stego images indicates improved resistance against statistical detection. Hence, the proposed method thus provides a reliable and more practical solution for secure data hiding in scenarios that demand both cryptographic strength and lowvisibility communication.

Recently, a shift in biomedical technology has occurred, specifically towards remote patient monitoring. This however, also increases the chances of cyberattacks like sensitive data theft, impersonation because of which many of systems face issues such as DDoS attacks, impersonation, MITM. Existing security mechanisms typically address these threats in isolation, focusing on either traffic filtering, authentication, or data-integrity protection. The proposed model aims to benefit the aforementioned networks by ensuring their safety as well as efficiency, using a comprehensive three-point system, where the three points are namely an authorization server, nodes and a database/middleware, each securely communicating with each other independent of the third point that ensures authenticity, efficiency, and resilience against the attack. The following model is has primary objective of minimizing the computational and communication costs while delivering the optimal efficiency with reduced overhead in authentication process and higher read-write efficiency under strict time and hardware constraints.

Intrusion detection has become increasingly challenging as modern networks grow larger, more distributed, and more privacy-sensitive. Traditional intrusion detection systems rely on centralized data collection, which not only increases communication cost but also raises concerns around data confidentiality. Federated learning offers an appealing alternative by enabling devices to train models locally, but most existing FL-based IDS frameworks still depend on a flat, single-layer architecture that struggles with non-IID data and limited scalability. To address these limitations, this study presents a three-tier federated learning architecture designed specifically for intrusion detection in distributed environments. Our framework organizes sensor nodes, cluster heads, and a central base station into a hierarchical structure that performs multi-level aggregation. This design reduces communication overhead, improves model stability, and better accommodates the diverse data patterns typically found in IoT and large-scale networks. Experiments conducted on the NSL-KDD dataset demonstrate that the proposed approach achieves stable performance, reaching an accuracy of approximately 78% while preserving data privacy. The ROC curve, confusion matrix, and convergence results further show that the hierarchical model outperforms a standard flat FedAvg setup in terms of consistency and robustness. Overall, the findings highlight the potential of multi-tier federated learning as a practical and efficient solution for intrusion detection in real-world distributed systems.

Telehealth systems have become increasingly important in modern healthcare delivery, but protecting sensitive patient data remains a critical challenge. Many existing systems struggle with vulnerabilities that compromise data integrity, confidentiality, and the ability to track who accessed what information and when. This framework presents a security approach that combines three key technologies: AES-256 encryption for keeping diagnostic information private, LSB (Least Significant Bit) steganography for embedding encrypted reports directly into medical images, and blockchain records for creating an audit trail. Our method works by first encrypting a diagnostic report, then hiding it within the pixels of a medical image using LSB embedding, and finally storing a digital fingerprint (hash) of the final image on a blockchain ledger. This integrated approach ensures that patient data remains confidential during transmission, the integrity of medical images is preserved and verifiable, and there is a permanent record of data sharing events. Unlike standalone security solutions that address only single aspects of the problem, our framework creates a unified system where encryption, steganography, and blockchain work together to establish a secure and trustworthy environment for telehealth communications.