Robust Image Watermarking Using Lifting Wavelet Transform and Convolutional Neural Network

Abstract

The research introduces an innovative approach to resilient image watermarking, where the fusion of the Lifting Wavelet
Transform (LWT) and the Convolutional Neural Network (CNN) forms the cornerstone. To refine the precision and resilience of the
watermarking process, the study integrates the Adaptive Gazelle Optimization Algorithm (AGOA) into the LeNet-5 model. This
integration lets AGOA fine-tune important hyperparameters like learning rate, batch size, and convolution kernel number without having to worry about the risks that come with doing it by hand. Overall, AGOA's main goal is to improve accuracy during the model training phase. This will make sure that the watermark can't be seen and is resistant to many possible attacks. Through the utilization of AGOA, the research endeavors to swiftly and efficiently pinpoint an array of optimal solutions, thus optimizing the watermarking process to its fullest potential. The performance of our AGOA-Improve-LeNet5 model is evaluated using 300 test images, with an average Peak Signal-to-Noise Ratio (PSNR) of 61.074 dB. The experimental outcomes demonstrate the effectiveness of our proposed methodology in achieving robust and high-quality image watermarking.