Robust and high capacity watermarking for image based on DWT-SVD and CNN

Digital watermarking technology is of great importance to protect the copyright of the owners and authenticate the security of the media. As digital images are vulnerable to some common attacks during transmission, it is very necessary to design a watermarking algorithm which can resist all kinds of common attacks. Nowadays, most of the watermarking algorithms published rely much on the locations of the pixels for watermark embedding, which results in less robustness. And some algorithms took some measures to increase the ability to resist attacks, but the measures taken limited the algorithm in watermark embedding capacity. In this paper, a robust watermarking algorithm based on convolution neural network (CNN) is proposed. We introduce discrete wavelet transform (DWT) technology and singular value decomposition (SVD) technology, to achieve the embedding process of watermark. The network is established in the spatial domain based on the pixels' relationships of watermark, host image and watermarked image. After that, The pixels of the watermarked image are lightly modified with the network. In addition, some attacks are taken to the watermarked image. Simulation shows that proposed algorithm has good performance, the watermark extracted can be clearly identified.