Robust Zero-Watermarking Technique Based on Do-Resnet and Adaptive Pelicon Optimization
Sambhaji Marutirao Shedole, Santhi Vaithiyanathan · International Journal of Signal and Imaging Systems Engineering · 2024
This research presents a revolutionary deep learning-based zero watermarking technology that appears to increase image security. Zero-watermarking was utilised to secure the copyright data of highly invisible images. The pretrained DO-ResNet model was developed to extract high-dimensional deep features from images. The deep features are selected using low-frequency coefficients of the discrete Fourier transform (DFT). Furthermore, employing an adaptive Pelican optimisation (APO) algorithm, the loss of result in the optimal area can be reduced. The experimental results demonstrate that the approach is robust, secure, invisible and it can obtain watermark information reliably. The proposed method handles geometric and common attacks more efficiently by automatically extracting high-dimensional, complex information from images. The obtained simulation results demonstrate that the USC-SIPI dataset, the proposed method performs better in terms of accuracy (99.32%), PSNR (51.62), and SSIM (97.35). For the MS-COCO dataset, the proposed method attains an accuracy (98.45%), PSNR (37.58%), and SSIM (93.25%).