Robust Digital Image Watermarking to Protect the Privacy of Medical Images

Abstract

The patient privacy, data integrity, and unauthorised usage and distribution of diagnostic information in the form of images made the medical image watermarking as a promising research area. Although the encryption techniques have helped to protect and transit the medical image data, still there is need of strong verifiable and protective method. This article proposes a framework known as Dimensional Phase Coherence Watermarking (DPCW). DPCW is a semi-blind, phase-based transform which embeds watermark into grayscale medical cover image by protecting diagnostic fidelity. There are four stages of DPCW: i) Binary watermark payload is expanded into seven-dimensional orthogonal phase-coefficient with the help of Fourier-basis lift. ii)  The resulting phase-tensor is projected into 2-dimensional complex-valued coherence matrix, whose phase angles are adapted to the local spatial-frequency energy of the cover image. iii) Multi-state superposition encoding is applied that distributes watermark bits across eight statistically independent coefficient groups. (iv) Lorenz attractor is applied that drives a deterministic temporal permutation, parameterised by ( = 10,  = 28,  = 8/3) with a private seed key . The watermark embedding process is done by  with  = 0.001. And the extraction requires only the seed key and the coherence-phase index; there is no need of original cover image. Experiment is conducted on 50 grayscale medical images of size 256 × 256, 8-bit, comprising 25 chest X-ray and 25 brain MRI slices from publicly available benchmark datasets. The watermark with a size of 64 × 64 binary payload show PSNR = 85.4 ± 1.2 dB and SSIM = 0.9997 ± 0.0001. The overall 23 types of attacks covering noise, compression, filtering, and geometric distortion were tested, the mean NC is 0.9783 and the overall robustness score is 95.99% is obtained. CPU embedding time is measured as 0.31 s per image. The results show that DPCW is a viable tool in assuring medical image integrity, offering a reproducible, reusable method to imperceptible watermark embedding and attack-resilient visual information protection.