Attack-Resilient Medical Image Watermarking Using Convolutional Neural Networks and Fuzzy Inference Decoding Mechanisms

Abstract

The rapid digitization of healthcare and widespread use of medical imaging systems have intensified concerns related to image security, authenticity, and patient data confidentiality. Conventional medical image watermarking techniques often fail to provide adequate robustness against diverse signal processing and geometric attacks while preserving diagnostic quality. To address this challenge, this work proposes an attack-resilient medical image watermarking framework that integrates Convolutional Neural Network (CNN)–based adaptive watermark embedding with fuzzy inference–based decoding. The proposed approach intelligently learns perceptually safe embedding regions and effectively manages uncertainty during watermark extraction under severe distortions. The methodology is evaluated using a publicly available multi-modal Medical Imaging (CT–Xray) Colorization New Dataset comprising 10,000 images, split into 60% training, 20% validation, and 20% testing. Experimental results demonstrate excellent robustness, achieving a Normalized Correlation (NC) of 1.0 and a Bit Error Rate (BER) of 0.0 under noise, compression, filtering, and geometric attacks. Classification performance using a hybrid CNN–RNN model attains an accuracy of 97.1%, outperforming standalone CNN (96.5%) and RNN (60.0%) models, confirming the effectiveness of the proposed framework.