Robust Blind Watermarking of Medical Images using ResNet-50 and Redundant Wavelet Transform

Abstract

Medical images are crucial in identifying and detecting certain disorders within healthcare necessities. Numerous secure and insecure networks communicate medical imaging, impacting critical clinical information. Despite the effectiveness of numerous medical image protection solutions, their robustness against complex intrusions has not received enough attention. This indicates the necessity to enhance medical image protection techniques against sophisticated attackers. To address this issue, we created a new strong blind watermarking method for medical images that uses ResNet-50 and the Redundant Wavelet Transform (RDWT) in a hybrid domain. We first divide the input image into 8x8 non-overlapping blocks for rapid computation and then feed it into the pre-trained ResNet-50 model to extract the stable feature vector accurately. RDWT transforms the obtained feature vector to generate the detailed coefficients LL, LH, HL, and HH. To improve security, we first encrypt the binary watermark (BW) using Arnold encryption to embed it in the selected LL sub-band. The pre-trained ResNet50 model, when combined with RDWT in a hybrid domain, effectively captures more intrinsic and localized features. The study's findings are highly promising, demonstrating the proposed method's effectiveness in robustness and invisibility. The embedded watermark can be extracted free of distortion. The extracted watermark appears genuine, demonstrating optimal BER and NC values. The BER values neared zero in nearly all attack scenarios, while the NC values approached one.