Real-time FPGA-based secure semantic adaptive watermarking system for high-resolution medical images
Аннотация
Medical image watermarking plays a critical role in securing patient data while maintaining diagnostic fidelity, but suffers from Initialization Vectors (IV) reuse vulnerabilities and falsification of critical regions. Hence, the EncryptAware Swin-Gray Watermark-Net is used within an FPGA-based real-time framework to solve the problems. Traditional encryption methods remain vulnerable to key-recovery attacks caused by improper reuse of IV in block cipher modes. Thus, Key-Derivation and Authenticated Encryption Mechanism (KD-AEM) is utilized, where hierarchical key derivation provides cryptographically independent session keys for each session, and SIV-based Authenticated Encryption prevents IV reuse and ensures the confidentiality of embedded patient data. Additionally, existing embedding strategies alter diagnostically sensitive regions due to a lack of semantic awareness, which fail to accurately quantify sub-voxel micro-patterns, leading to faulty importance weighting. To accurately protect diagnostically sensitive regions, Swin-Gray-Level Firefly Wavelet Transform (Swin-GL-FWT) is introduced, which extracts voxel-level semantic importance maps, detects sub-voxel micro-patterns, and enhances recognition of low-amplitude, high-frequency variations. Finally, Firefly-Discrete Wavelet Transform (F-DWT) dynamically optimizes embedding strength and location based on these maps, thus balancing imperceptibility, and prevents over-embedding. Experimental results show that the proposed model beats the existing models in terms of watermarking with a higher SSIM of 0.995 and PSNR of 92 dB.