Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Imperceptible watermark attack method based on residual learning, storage medium and electronic device

A watermarking attack and residual technology, applied in image watermarking, image data processing, instruments, etc., can solve problems such as interference and immature attack system

Pending Publication Date: 2021-09-10
DALIAN MARITIME UNIVERSITY
View PDF0 Cites 10 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

In 2020, Nam et al. proposed a watermarking attack network. They realized that the existing attack schemes still cannot be used as a benchmark for testing the robustness of watermarking schemes, and there are many problems
[0006] Although the current few research results show that deep learning technology can be used as a new watermark attack method to interfere with watermark extraction and at the same time ensure the image quality of the watermark, the attack system is still not mature enough, and most of the current watermark attack schemes are based on improving the watermark. The quality of the image (PSNR, SSIM, etc.) is the goal, ignoring the problem of watermark extraction (BER, Bit Error Rate)

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Imperceptible watermark attack method based on residual learning, storage medium and electronic device
  • Imperceptible watermark attack method based on residual learning, storage medium and electronic device
  • Imperceptible watermark attack method based on residual learning, storage medium and electronic device

Examples

Experimental program
Comparison scheme
Effect test

Embodiment approach

[0080] S2. Based on the constructed watermark attack model, design a loss function; including selecting an appropriate number of feature extraction blocks according to the embedding area of ​​the watermark information to extract a feature map containing a large amount of watermark information; introducing a residual learning mechanism, by learning the watermark The difference between the image and the unwatermarked image is used to improve the convergence speed and learning ability of the watermark attack model. During specific implementation, as a preferred embodiment of the present invention, it specifically includes the following steps:

[0081] Step 1.1, the watermark attack model extracts the low-frequency feature image representing the watermark information through the feature extraction block, and in order to destroy the watermark information in the watermarked image, the watermarked image is subtracted from the low-frequency feature image to obtain a residual image; nam...

Embodiment 1

[0118] In order to verify the effectiveness of the inventive method, a simulation experiment was carried out, and the following experimental results were given, and the experimental results were analyzed simultaneously, specifically as follows:

[0119] 1. Embedding area for watermark information:

[0120] The method of color image robust watermarking based on quaternion exponent moments embeds watermark information into the specific area of ​​the original image without watermark, which determines the structural design of the watermark attack model. If the watermark information is embedded in the high-frequency region of the original image without watermark, then the feature extraction module of the watermark attack model is mainly used to extract the high-frequency information of the watermarked image, and the residual image (including the watermarked image and the extracted feature image The watermark information that is poorly done) will be destroyed to the greatest extent;...

Embodiment 2

[0126] 2. For the depth of network architecture

[0127] In order to verify the impact of the depth of the network architecture on the attack capability of the model, the feature extraction blocks of this embodiment are respectively selected as 7, 11, 15 and 20, and 25 watermarked images are randomly selected from the test set to calculate the bit error of extracting the watermark information rate, the experimental results such as Figure 5 shown, from Figure 5 It can be seen from the experimental results that when the number of layers N of the feature extraction block is 2, the average bit error rate of watermark information extraction is 0.11716, indicating that the watermark attack model has a certain destructive ability; while the number of layers N of the feature extraction block When 7 is selected, the average bit error rate of watermark information extraction is as high as 0.13885. It can be concluded that as the depth of the model increases, the attack capability of ...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention provides an imperceptible watermark attack method based on residual learning, a storage medium and an electronic device. The method comprises the steps: constructing a watermark attack model based on a convolutional neural network, carrying out the end-to-end nonlinear learning between a watermark-containing image and a watermark-free image, mapping the watermark-containing image to the watermark-free image, and carrying out watermark attack; selecting a proper number of feature extraction blocks according to the embedded area of the watermark information to extract a feature map containing the watermark information; introducing a residual learning mechanism to improve the convergence speed and learning ability of a watermark attack model, and improving the imperceptibility of an attacked image by reducing the difference between a residual image and the watermarking-free image; and constructing a data set for training a watermark attack model according to a DIV2K2017 super-resolution data set and a robust color image watermark algorithm based on a quaternion exponential moment. The watermark attack model provided by the invention can attack a robust watermark algorithm at a high bit error rate on the premise of not destroying the visual quality of a watermark-containing image.

Description

technical field [0001] The present invention relates to the technical field of digital watermarking, in particular to an imperceptible watermarking attack method based on residual learning, a storage medium and an electronic device. Background technique [0002] With the continuous development of Internet technology, information acquisition has become more and more convenient, followed by the unrestricted storage and transmission of massive information on the network. How to effectively protect information security is a key issue that needs to be solved urgently, and it has always been one of the most important topics in scientific research. Digital watermark technology is the key technology of digital image copyright protection. By embedding some identification information (that is, digital watermark) into the image to be protected, the purpose of confirming the copyright of the image is achieved. At present, the research on digital watermarking technology mainly focuses o...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
IPC IPC(8): G06T1/00G06N3/04
CPCG06T1/005G06N3/045Y02T10/40
Inventor 王兴元李琦王晓雨咸永锦高锁闫晓鹏
Owner DALIAN MARITIME UNIVERSITY
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products