通过对抗性净化和微调来提高UDC图像恢复的稳定性
Wenjie Dong1, Zhenbo Song1, Zhenyuan Zhang1
1School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing 210094, China.
Sensors (Basel, Switzerland)
|September 19, 2025
概括
本研究介绍了UDC图像修复模型对抗对抗攻击的新防御方法. 我们的框架显著提高了攻击后的模型稳定性和图像恢复质量.
科学领域:
- 计算机视觉 计算机视觉
- 机器学习 机器学习
- 图像处理 图像处理
背景情况:
- 显示屏下摄像头 (UDC) 图像恢复模型容易受到敌对攻击.
- 现有的防御措施无法充分解决这一特定的漏洞.
研究的目的:
- 评估基于深度学习的UDC图像恢复模型对各种对抗性攻击的稳定性.
- 提出和验证一个新的防御框架,以提高UDC模型的弹性.
主要方法:
- 使用白盒和黑盒攻击 (例如,PGD,C&W) 系统评估七种最先进的UDC模型.
- 发展两阶段防御:基于扩散的对抗性净化和高效的微调.
- 对拟议的防御框架进行了广泛的实验验证.
主要成果:
- 在对抗性攻击下的UDC模型中观察到显著的性能下降 (例如,DISCNet PSNR从35.24下降到15.16).
- 拟议的防御框架成功地恢复了性能,DISCNet的PSNR在PGD攻击下反弹到32.17.
- 在防御后,UFormer实现了针对LPIPS引导攻击的改进PSNR (19.71比17.38基线).
结论:
- UDC图像恢复模型非常容易受到对抗性干扰的影响.
- 拟议的基于扩散的净化和微调框架为对抗攻击提供了有效的防御.
- 防御框架增强了UDC模型的弹性,而不会影响图像恢复的保真性.
相关概念视频
Survival Tree
389
Survival trees are a non-parametric method used in survival analysis to model the relationship between a set of covariates and the time until an event of interest occurs, often referred to as the "time-to-event" or "survival time." This method is particularly useful when dealing with censored data, where the event has not occurred for some individuals by the end of the study period, or when the exact time of the event is unknown.
Building a Survival Tree
Constructing a...
Building a Survival Tree
Constructing a...
389
Detergent Purification of Membrane Proteins
6.3K
Detergents are used to purify the integral proteins of the membrane. The hydrophobic portion of the detergent can replace membrane phospholipids while solubilizing the membrane proteins. When detergent monomers reach a specific concentration in a solution called critical micelle concentration (CMC), they form micelles. Above CMC, the concentration of the detergent monomers remains in equilibrium with the micelle. The number of detergent monomers present in the CMC varies for each detergent, and...
6.3K
Upsampling
588
Managing signal sampling rates is essential in digital signal processing to maintain signal integrity. A decimated signal, characterized by a reduced frequency range due to its lower sampling rate, can be upsampled by inserting zeros between each sample. This upsampling process expands the original spectrum and introduces repeated spectral replicas at intervals dictated by the new Nyquist frequency. To refine this zero-inserted sequence, it is passed through a lowpass filter with a cutoff...
588
