Robust Multi-watermarking Algorithm Based on DarkNet53
Dekai Li, Jingbing Li, Uzair Aslam Bhatti · 2023
This chapter introduces a novel method to address the robustness problem in multi-watermarking algorithms for medical images. The approach combines discrete cosine transform (DCT) with DarkNet53 convolutional neural network to enhance the algorithm&s;s robustness. To achieve this, the algorithm utilizes a pre-trained DarkNet53 network for transfer learning. The original softmax layer and classification layer are substituted with a fully connected layer and a regression layer, respectively. This modification converts the classification network into a regression network, generating 32 features as output. The medical image is then processed through the network, and the distinctive features are extracted by the fully connected layer. These features are subsequently fused with DCT and perceptual hashing to further strengthen the algorithm&s;s robustness. The experimental results demonstrate the accurate differentiation of various medical images by the proposed algorithm, as well as the efficient recovery of the original information from the encrypted multi-watermarked data under traditional and geometric attacks. Moreover, the algorithm exhibits superior robustness and invisibility compared to existing approaches.