Robust watermarking requires finding invariant features under multiple attacks to ensure correct extraction. Deep learning has extremely powerful in extracting features, and watermarking algorithms based on deep learning have attracted widespread attention. Most existing methods use 3 × 3 small kernel convolution to extract image features and embed the watermarking. However, the effective perception fields for small kernel convolution are extremely confined, so the pixels that each watermarking can affect are restricted, thus limiting the performance of the watermarking. To address these problems, we propose a watermarking network based on large kernel convolution and adaptive weight assignment for loss functions. It uses large-kernel depth-wise convolution to extract features for learning large-scale image information and subsequently projects the watermarking into a highdimensional space by 1 × 1 convolution to achieve adaptability in the channel dimension. Subsequently, the modification of the embedded watermarking on the cover image is extended to more pixels. Because the magnitude and convergence rates of each loss function are different, an adaptive loss weight assignment strategy is proposed to make the weights participate in the network training together and adjust the weight dynamically. Further, a high-frequency wavelet loss is proposed, by which the watermarking is restricted to only the low-frequency wavelet sub-bands, thereby enhancing the robustness of watermarking against image compression. The experimental results show that the peak signal-to-noise ratio (PSNR) of the encoded image reaches 40.12, the structural similarity (SSIM) reaches 0.9721, and the watermarking has good robustness against various types of noise.

Single image deraining methods have been extensively studied for its ability to remarkably improve the performance of computer vision tasks in rainy environments. However, most existing rain removal methods still have two major drawbacks which are hindering the technology development. First, the rain streaks are seriously coupled with the background information in a single rainy image, which leads to incorrect identification of rain streaks by many methods and further makes the loss of texture details in the rain removal results. Second, they spend excessive computational cost, which is not conducive to practical applications. To address these issues, a progressive separation network (PSN) is proposed by decomposing the rain removal task into two stages, the bilateral grid learning stage and the joint feature refinement stage, from a novel perspective. The bilateral grid learning stage is designed to expand the distance between the rain streaks and the background information while preserving the image edge details to guide the subsequent refinement. For the joint feature refinement stage, a dual‐path interaction module is constructed to dynamically and gradually decouple the rain streak content and the intermediate features of the clear image details. In addition, an activation‐free feature refinement block is designed to further improve the computational efficiency by removing or replacing the activation function without loss of accuracy. Extensive experiments on synthetic and real datasets show that PSN outperforms state‐of‐the‐art rain removal methods in terms of quantitative accuracy and subjective visual quality. Furthermore, competitive results are derived by extending PSN to the defogging task.

No abstract