DEVELOPMENT OF DATA HIDING METHODS FOR CONTENT PROTECTION IN MEDICAL DOMAIN

Abstract

Reversible Data Hiding (RDH) has emerged as a vital research area within information security, driven by the growing demand for secure communication that preserves data confiden- tiality while ensuring perfect recovery of the original cover image. Recognizing its significance across sensitive domains such as medical imaging, cloud storage, and digital forensics, this thesis presents an extensive investigation into contemporary RDH techniques. Building upon insights derived from a comprehensive review of state-of-the-art methods, the work identifies key research gaps and proposes novel methodologies aimed at enhancing embedding efficiency, prediction accuracy, and visual fidelity in reversible data hiding. The thesis begins with an in-depth exploration of RDH by exploring major techniques and advancements within the field in the last decade and so. It systematically analyzes the progres- sion of RDH methodologies, highlighting key contributions, emerging trends, and research fron- tiers identified through an extensive survey of scholarly works indexed in reputable databases such as the Web of Science (WoS). This review not only consolidates the existing body of knowledge but also identifies prevailing challenges, unresolved research issues, and potential avenues for future investigation in the domain of RDH. In the second part of the research, a high-capacity RDH method integrating contrast en- hancement and brightness preservation is proposed for medical images. The method divides the image into Region of Interest (ROI) and Non-Region of Interest (NROI), applying region- specific embedding strategies that align with their distinct characteristics. A novel prepro- cessing technique is further introduced for the ROI, efficiently handling areas with low Pixel Concentration Ratio (PCR) to generate additional vacant bins for embedding. This approach significantly enhances embedding capacity while improving contrast and visual quality without compromising the inherent brightness of the medical image. The third contribution introduces an advanced RDH method for color images employing a Convolutional Neural Network Convolutional neural network (CNN)–based predictor. To ad- dress the limited dependency range in conventional RDH methods, a Self-Attention CNN (SA- CNN) predictor is designed to capture both local and global spatial dependencies. A novel error adjustment mechanism further leverages the inter-channel correlation among RGB components, v resulting in superior reconstruction quality and increased embedding performance. The fourth part of the work presents a Two-Stage Interpolation-Based Reversible Data Hid- ing Framework that integrates attention-driven prediction to enhance embedding efficiency. To overcome the limitations of conventional interpolation techniques, the proposed approach com- bines bicubic interpolation with a highly accurate deep-learning predictor. A novel multi-head attention–based U-Net model, termed UMANet, is introduced to capture a broader spatial con- text, yielding improved interpolation accuracy and embedding performance. The fifth contribution focuses on Reversible Data Hiding in Encrypted Images (RDHEI), adding an additional layer of security over traditional RDH. A novel SCAM-Net predictor is proposed, equipped with a multiscale extraction module for capturing both fine and coarse feature details. The extracted features are refined using a Convolutional Block Attention Module (CBAM) that leverages channel and spatial attention mechanisms. This architecture achieves highly precise image prediction and outperforms existing state-of-the-art methods, leading to superior embedding performance and robust data concealment in encrypted environments. Collectively, the research contributes a comprehensive suite of RDH methodologies that advance the state of the art in prediction accuracy, embedding efficiency, and visual fidelity, reinforcing the role of intelligent predictive modeling in secure image-based communication and storage.