Improving the Security of Medical Imaging via DFT-Based Reversible Watermarking and Deep Learning-Based Zero-watermarking

This paper introduces a hybrid image verification and authentication technique, integrating reversible-watermarking with zero-watermarking to improve the security of medical images. In zero-watermarking, a convolutional neural network extracts specific image features and merges them with a patient's image to create a master share. To increase the robustness, the QR code is fused with the master share. The QR code is embedded into the image using a reversible watermarking technique into Regions of Non-Interest detected by the K-means algorithm, ensuring the optimal region for the QR code embedding into the middle-frequency coefficients of the Discrete Fourier Transform. The experimental results validate the robustness of the proposed scheme despite applying different ge-ometric transformations and image processing distortions to the watermarked image. The results demonstrate that our method preserves image quality and recovers the watermark efficiently. The obtention of a low bit error rate and high normalized cross-correlation values evidences the efficiency.