Jun 2026· IEEE Transactions on Emerging Topics in Computing· Vol abs/2606.29417, pp. 1-14· 0 citations· 46 references
Computer Science
TL;DR
The Bit-ViP scheme produces secure, usable images by incorporating an innovative end-to-end obfuscation function, and would contain non-invertible noise, making it hard for an adversary to reconstruct the original image.
Abstract
The unprecedented growth of computer vision applications, such as surveillance systems and social media, raises security and visual privacy concerns, especially when data is stored on cloud servers. Image obfuscation offers a way to preserve visual privacy while maintaining an adequate level of usability; thus, it has been a topic of great interest in recent years. However, prior obfuscation schemes are either vulnerable to malicious attacks, such as model inversion to reconstruct original images from obfuscated images, or generate non-trainable obfuscated images, making them unusable for achieving reasonable accuracy. This paper proposes a novel bit-plane-based image obfuscation scheme, {\em Bit-ViP}, to preserve visual privacy for image-based recognition tasks. The Bit-ViP scheme produces secure, usable images by incorporating an innovative end-to-end obfuscation function. While doing so, the obfuscated image would contain non-invertible noise (generated by Lorenz's chaotic system and differential privacy), making it hard for an adversary to reconstruct the original image. We conduct extensive experiments on two popular activity recognition datasets, namely UCF101 and HMDB51, to validate the effectiveness of Bit-ViP. In the face of attacks on reconstruction, pixel frequency, information entropy, and pixel inter-correlation, we present a rigorous security analysis demonstrating tangible improvements over existing schemes.
Modern cloud platforms now carry a substantial share of the world's sensitive digital traffic, and defending that traffic
calls for more than hiding its meaning - an intercepted ciphertext is still visibly a target worth attacking. Encryption alone
protects content; it does nothing to disguise the fact that a secret is being sent at all. This paper proposes VisionStego, a duallayer security architecture that pairs symmetric-key encryption with an artificial-intelligence-guided steganographic embedding
stage, so that cloud-hosted data is protected in both substance and appearance. The secret payload is first reduced in size
through wavelet-based compression, then encrypted with a shared symmetric key, and finally concealed inside a cover image at
positions chosen by a trained Convolutional Neural Network (CNN) rather than by a fixed or pseudo-random rule. The network
scores each candidate pixel for embedding suitability using edge strength, local variance, and texture complexity, concentrating
modification in regions where it is least visible to the eye and least anomalous to statistical steganalysis. Tested on the Lena,
Baboon, and Peppers benchmark images, the resulting stego images reach a Peak Signal-to-Noise Ratio (PSNR) of up to 46.2
dB and a Structural Similarity Index (SSIM) of 0.985, exceeding conventional LSB, adaptive LSB, and wavelet-based baselines
on every image and every metric tested. These findings indicate that VisionStego offers a practical route to the combined
concealment and confidentiality that neither cryptography nor steganography can deliver in isolation.
R. Saxena, Priti Maheshwary· International Journal for Re...· 0 citations
Federated learning enables collaborative model training while keeping data locally at each client; however, recent studies have shown that training data can be reconstructed from shared model updates. To address this issue, this paper proposes a dual obfuscation method that enhances robustness against image restoration attacks by jointly obfuscating updated information and training images. The proposed method combines a robustness enhancement technique based on random binary weights, which randomly sets a portion of gradient elements to zero, with an image encryption technique. These techniques provide complementary protection by reducing the amount of original gradient information available to an attacker and the visual interpretability of reconstructed images, respectively. Furthermore, the image encryption technique allows independent keys to be used for each client and each image, avoiding explicit key sharing. Experimental results on an image classification task using a Vision Transformer (ViT) show that the proposed method reduces the visual information recovered by Attention Privacy Leakage (APRIL) under the evaluated settings without causing additional degradation in classification performance beyond that caused by image encryption. Although the proposed combination does not provide an absolute security guarantee, the results demonstrate the potential benefit of combining gradient modification and image encryption for privacy-enhanced federated learning.
Yukihiro Itabashi, Hiroto Sawada, Mare Hirose et al.· 0 citations
Identity document (ID) authentication relies on the structural integrity of complex, high-frequency security patterns. However, advanced Generative AI models can now inject localized, high-fidelity manipulations, creating deceptive attacks that bypass standard verification. Training robust image forensic models to detect these anomalies is hindered by privacy regulations, forcing reliance on synthetic templates lacking the intricate visual patterns of real IDs. To bridge this domain gap, we introduce FakeIDet3-DB, the first comprehensive database of digital manipulations on real, government-issued IDs. FakeIDet3-DB encompasses classical (e.g., copy-move) and Generative AI-driven manipulations (e.g., face-swapping, inpainting) enhanced with advanced image refinement procedures to suppress visual artifacts. In addition, to comply with strict data protection regulations (e.g., GDPR), we adopt a recently-proposed framework based on patches. In order to maximize forensic utility, we formulate privacy-aware patch extraction from a real ID as a geometrically constrained image processing problem. We propose PACE, a Pseudo-Anonymized Contextual patch Extraction algorithm, which leverages Integral Image mapping and distance-driven Non-Maximum Suppression (NMS). PACE efficiently contours anonymization masks that prevent Personally Identifiable Information (PII) leakage while maximizing semantic density in peri-censorship regions, yielding almost 5.2M patches extracted from more than 6.4K images from real/fake IDs. Furthermore, an extensive evaluation of the proposed FakeIDet3-DB is performed using state-of-the-art models, showcasing they all struggle to detect and locate attacks coming from generative and classic techniques (32.45\% EER in detection and 83.48\% AUC-ROC in localization).
Muñoz-Haro Javier, Teruel Andres, Tolosana Ruben et al.· 0 citations
The increased rate of cyber threats such as fraud and other forms of attack, including phishing, malware, and denial-of-service attacks, has led to a growing demand for secure methods to ensure the security of sensitive information that is transferred among users. Video steganography has become a very important element in dealing with these security issues, as communication has become more dependent on multimedia content. Hiding information is now an area that is being developed very fast with the introduction of Deep Learning (DL)-based steganography methods. This framework introduces Vid_Steg_DenseUNet, a video steganography method which efficiently extracts multi-level features using edge-preserving U-Net and DenseNet through Invertible Neural Networks (INN) interactions. The main objectives of this approach include improving the perceptual fidelity of the stego video and enhancing the quality of the recovered secret image. These improvements are evaluated using objective imperceptibility metrics through pixel-level and structural analyses, such as PSNR and SSIM. The model was trained on the DIV2K images and Ultra Video Group (UVG) videos. The proposed model hides a secret color image in each video frame, and its performance is assessed using quantitative difference metrics. Comparative results with existing state-of-the-art methods, such as recent DL-based video steganography frameworks, show that this end-to-end embedding approach delivers promising performance in terms of human visual accuracy, with a PSNR score of 37.054 and an SSIM score of 0.9754. The results also demonstrate that the model achieves enhanced security and high resistance to detection techniques.
Noor Fahem Sahib, S. Hashem, E. F. Naser· Baghdad Science Journal· 0 citations
Recently, in the science of data hiding, most research papers propose several techniques to conceal data inside the images and ensure secure mediums such as text, images, audio, and videos with preserving its quality. Researchers have been interested in the last decade of these techniques in image steganography and cryptography, this research proposed a system that uses multiple layers of security in which steganography and cryptography are together to enhance security. This study aims to present a new efficient technique for hiding a gray image inside a blue layer of color image by combining the genetic algorithm (GA) with the Least Significant Bit (LSB) approach to optimal solution permutation for embedding pixel assortment of the image where data is to be concealed. This technique offers immovability differential attacks that are evaluated by several performance metrics. According to experimental findings, the stego and the cover image are visually indistinguishable. PSNR, MSE, and SSIM are used as measurement matrices. Our aim passed a successful test of robustness against experimental analysis.
Dena Abu Laila, Ziad Dawahdeh, Amer Alqutaesh et al.· Journal of Cyber Security an...· 0 citations
Spatial domain image steganography has emerged as one of the most widely adopted information-hiding techniques due to its simplicity, high embedding capacity, low computational complexity, and ability to preserve the visual quality of digital images. It plays a significant role in secure communication by concealing confidential information within digital images, thereby protecting sensitive data from unauthorized access in applications such as healthcare, military communication, banking, cloud computing, digital forensics, and multimedia systems. Despite these advantages, spatial domain techniques face several challenges, including vulnerability to steganalysis, limited robustness against image processing operations, and the trade-off between embedding capacity and imperceptibility. This paper presents a comprehensive review of spatial domain image steganography by examining its historical development, fundamental concepts, classification, and major techniques, including Least Significant Bit (LSB), Adaptive LSB, Pixel Value Differencing (PVD), Pixel Indicator Technique (PIT), Optimal Pixel Adjustment Process (OPAP), edge-based methods, and other adaptive spatial approaches. The reviewed techniques are comparatively analyzed based on embedding capacity, imperceptibility, robustness, computational complexity, and practical applicability, and are further illustrated through a quantitative case study that evaluates PSNR, SSIM, and MSE for LSB substitution on a standard test image. The study identifies existing research gaps and highlights future research directions aimed at improving robustness, visual quality, and embedding capacity. By providing a structured and critical synthesis of existing literature, this review serves as a valuable reference for researchers, academicians, and practitioners working in the field of digital image security and information hiding.
Keywords: Least Significant Bit (LSB); Pixel Value Differencing (PVD); Information Hiding; Digital Image Security; Steganalysis
Niranjana K. S., Sidha P. P., A. G et al.· International Journal of Tec...· 0 citations