Secure Blind Medical Image Watermarking Using Hybrid Feature Extraction Techniques

Watermarking offers great potential for medical images by embedding identifiable information that ensures secure and authenticated sharing of patient data while maintaining both integrity and diagnostic quality. In this paper, we present an innovative framework for blind medical image watermarking that harnesses advanced feature extraction techniques, including K-Means clustering, BRISK (Binary Robust Invariant Scalable Key-points), GFTT (Good Features to Track), and chaotic systems algorithms. We conducted extensive experiments on the Ocular Disease Intelligent Recognition (ODIR) dataset, focusing specifically on Retinal Optical Coherence Tomography (OCT) images. The results highlight the framework's ability to preserve image quality and diagnostic utility, with minimal perceptual impact, as evidenced by strong evaluation metrics: a Mean Squared Error (MSE) of 0.0001, a Peak Signal-to-Noise Ratio (PSNR) of 85.27, a Structural Similarity Index Measure (SSIM) of 0.9999, a Normalized Correlation Coefficient (NC) of 1, and a Universal Average Change Index (UACI) of 0. Overall, this framework represents a significant step forward in the secure sharing of medical images, ensuring that patient data remains protected and accessible without compromising diagnostic fidelity.