Robust watermarking with PSO and DnCNN

One of the challenges that Singular Value Decomposition (SVD)-based algorithms in watermarking face is the false positive problem while maintaining important parameters in watermarking, especially in different attacks. False positives occur when the watermark is placed on the S component of the SVD transformation of the host image. The proposed solutions have problems such as the low-quality watermarked image and its lack of stability against various attacks. To solve such problems, low-frequency bands of the host image, which are obtained by transforming the Contourlet obtained by taking the SVD transform and separating the S component are embedded by multiplying a parameter called delta obtained from the PSO (Particle swarm optimization) algorithm in the watermark image, and the embedding stage ends at this point. The result of this method is a relatively improved NC value, especially for the salt and pepper attack, which is compared with different articles in the results section. In the extraction section, the contourlet transform is taken from both the host image and the watermark, and the results are expressed with and without the denoising convolutional neural network algorithm (DnCNN). The values obtained for the false positive problem in the results section show that this scheme, while not having such a problem, is also resistant to various attacks.