Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Picking watermarks from noise (PWFN): an improved robust watermarking model against intensive distortions

Sijing Xie, Chengxin Zhao, Nan Sun, Wei Li, Hefei Ling

TL;DR

The paper addresses the limited robustness of deep-learning watermarking under strong distortions by introducing PWFN, which places a denoise module between the noiser and decoder and employs an SE-based encoder to enhance watermark–image coupling. The approach transforms robust watermark extraction into a denoising recovery problem and optimizes a joint loss over encoder, denoiser, decoder, and discriminator, enabling end-to-end training. Key contributions include the denoise module for restoring degraded signals and the SE-encoder for improved coupling, with ablation showing their effectiveness; results demonstrate competitive performance and superiority at higher noise intensities, along with cross-dataset generalization. This framework offers practical enhancements for robust watermarking in realistic distortion scenarios and can be extended to existing watermarking networks to improve robustness and visual quality.

Abstract

Digital watermarking is the process of embedding secret information by altering images in an undetectable way to the human eye. To increase the robustness of the model, many deep learning-based watermarking methods use the encoder-noise-decoder architecture by adding different noises to the noise layer. The decoder then extracts the watermarked information from the distorted image. However, this method can only resist weak noise attacks. To improve the robustness of the decoder against stronger noise, this paper proposes to introduce a denoise module between the noise layer and the decoder. The module aims to reduce noise and recover some of the information lost caused by distortion. Additionally, the paper introduces the SE module to fuse the watermarking information pixel-wise and channel dimensions-wise, improving the encoder's efficiency. Experimental results show that our proposed method is comparable to existing models and outperforms state-of-the-art under different noise intensities. In addition, ablation experiments show the superiority of our proposed module.

Picking watermarks from noise (PWFN): an improved robust watermarking model against intensive distortions

TL;DR

The paper addresses the limited robustness of deep-learning watermarking under strong distortions by introducing PWFN, which places a denoise module between the noiser and decoder and employs an SE-based encoder to enhance watermark–image coupling. The approach transforms robust watermark extraction into a denoising recovery problem and optimizes a joint loss over encoder, denoiser, decoder, and discriminator, enabling end-to-end training. Key contributions include the denoise module for restoring degraded signals and the SE-encoder for improved coupling, with ablation showing their effectiveness; results demonstrate competitive performance and superiority at higher noise intensities, along with cross-dataset generalization. This framework offers practical enhancements for robust watermarking in realistic distortion scenarios and can be extended to existing watermarking networks to improve robustness and visual quality.

Abstract

Digital watermarking is the process of embedding secret information by altering images in an undetectable way to the human eye. To increase the robustness of the model, many deep learning-based watermarking methods use the encoder-noise-decoder architecture by adding different noises to the noise layer. The decoder then extracts the watermarked information from the distorted image. However, this method can only resist weak noise attacks. To improve the robustness of the decoder against stronger noise, this paper proposes to introduce a denoise module between the noise layer and the decoder. The module aims to reduce noise and recover some of the information lost caused by distortion. Additionally, the paper introduces the SE module to fuse the watermarking information pixel-wise and channel dimensions-wise, improving the encoder's efficiency. Experimental results show that our proposed method is comparable to existing models and outperforms state-of-the-art under different noise intensities. In addition, ablation experiments show the superiority of our proposed module.
Paper Structure (18 sections, 10 equations, 4 figures, 3 tables)

This paper contains 18 sections, 10 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Deep learning Digital Watermarking
  4. Image Denoising
  5. Methods
  6. Encoder
  7. Noiser Layer
  8. Denosier
  9. Decoder
  10. Discrimator
  11. Total Loss
  12. Experiment and Results
  13. Experiment Setting
  14. Datasets
  15. Metrics
  16. ...and 3 more sections

Figures (4)

  • Figure 1: Our approach embeds the given message into the image with enhanced robustness and better visual quality. (a)original images,(b)encoded images,(c) residual images.
  • Figure 2: Mixes the pixel-wise and channel-wise information of the watermark, where $F_{sq}(\cdot)$, $F_{ex}(\cdot)$ and $F_{scale}(\cdot)$ means the squeezing, excitation, and scaling operation respectively.
  • Figure 3: Overview of the proposal network framework. The encoder generates the encoded image using the image and watermark information. The encoded image is then subjected to common noise processing to create a distorted image that resists noise. Then, the denoiser module reduces the distortion caused by the noise in the coded image. Finally, the decoder extracts the embedded watermark information.
  • Figure 4: Robustness comparison with baseline network under different noise levels.(a)Jpeg,(b)Crop,(c)Dropout,(d)Gaussian Blur,(e)Resize, (f)Gaussian Noise.