Secure Medical Image Authentication Using Zero-Watermarking based on Deep Learning Context Encoder

Abstract

Zero-watermarking is a robust and lossless technique for digital image security, copyright protection, and content authentication. This paper introduces a novel zero-watermarking scheme for medical image authentication based on deep learning. The proposed approach leverages a neural network based on the Context Encoder to extract distinctive features from the image, enhancing the method. The training of the neural model increases robustness. The watermark consists of a halftone image of the patient's face, serving as a unique identifier for medical study. By revealing the watermark, medical professionals can verify the correspondence between the imaging study and the patient. Therefore, an XOR operation merges the watermark sequence and the extracted features. The proposed method offers continuous image protection, safeguarding sensitive medical data. Extensive experiments demonstrate the technique's robustness against various attacks, including geometric transformations (scaling, cropping, resizing, rotation) and image processing manipulations (filtering, blurring, JPEG compression, and noise addition). The detection watermark process achieves a low bit error rate and a high normalized cross-correlation, validating the method's robustness and effectiveness. The deep neural network improved the robustness of the presented zero-watermarking scheme making it suitable for practical applications in medical data security and integrity.