Reversible watermarking for medical images using ROI-based tamper localization and recovery
Aulia Arham, Teguh Bharata Adji, Syukron Abu Ishaq Alfarozi, Hanung Adi Nugroho · Franklin Open · 2026
Medical image watermarking has received increasing attention in recent years for securing patient information and ensuring image integrity. To prevent diagnostic errors, data embedding must be fully reversible, and the watermarking scheme must detect and restore tampered regions. This paper proposes a reversible watermarking technique that supports high-capacity data embedding while preserving diagnostic quality and enabling authentication and recovery. The image is divided into three regions: the ROI, where patient records are reversibly embedded using a multi-layer difference expansion scheme; the RONI, where semi-fragile 2D-DE embedding enables tamper localization and recovery; and the border region, which stores auxiliary information using simple, non-robust LSB encoding that does not affect diagnostic content. Experimental results on MRI, ultrasound, and X-ray images show that the method can embed up to 10 kB of patient data using only 12% of the image as ROI, while maintaining high visual quality with average PSNR values above 38 dB and SSIM values close to 1.00. The watermarked images are perfectly reconstructed after extraction, except for the non-reversible border. A clear security and threat model is defined, assuming attackers may manipulate images, add mild noise, or structurally alter content. Under this model, the scheme successfully detects and localizes tampered regions through hash comparison and block analysis, and recovers corrupted ROI areas with quality improvements of up to +28.14 dB. The proposed method ensures full reversibility and reliable extraction of patient information when the ROI is not tampered, while lossy compression–based recovery provides visually meaningful restoration for tampered regions.