Robust Watermarking with PSO and DnCNN
Ali Amiri, Bahram Kimiaghalam · Research Square · 2024
Abstract The NC value obtained for our scheme is very good against jpeg attacks and the results are much better than the compared articles. Also, the NC value obtained for the salt and pepper attack is better than the compared articles. In this paper, an algorithm for multiple watermarking in gray scale images based on contourlet and SVD transforms is presented. Placing a watermark on the S component of the SVD transform of host image is associated with a false positive problem. Different solutions have been proposed, but these solutions themselves face problems such as determining the amount of scaling factor for different images, the undesirable values obtained for robustness to attacks, and most importantly, the low quality of the watermarked image. In the proposed scheme, to solve these problems in the embedding stage, we multiply the Δ value by the host image and multiply the component S obtained from the SVD transform and the SVD is taken from it again using the analysis of the host image by the contourlet and taking the SVD transform from it, and with the help of the PSO transform to find the scaling factor. In the extraction section, contourlet transform is taken from both the host and watermark images and the results are expressed with and without DnCNN. Attack robustness has also been compared with other articles using different images and applying different attacks. The values obtained for false positive in the results section show that this scheme does not have such a problem and is also robust to various attacks.