Enhancing robustness of weather removal: preprocessing-based defense against adversarial attacks

Vladimir Frants; Sos Agaian

doi:10.1117/12.3019864

7 June 2024 Enhancing robustness of weather removal: preprocessing-based defense against adversarial attacks

Vladimir Frants, Sos Agaian

Proceedings Volume 13033, Multimodal Image Exploitation and Learning 2024; 130330L (2024) https://doi.org/10.1117/12.3019864
Event: SPIE Defense + Commercial Sensing, 2024, National Harbor, Maryland, United States

Conference Poster

Abstract

Recent progress in deraining and dehazing methods has dramatically enhanced image quality in bad weather. However, these methods are vulnerable to adversarial attacks, severely compromising their effectiveness. Traditional defenses like adversarial training and model distillation necessitate significant retraining, hindering their real-world application due to high computational costs. To address these limitations, we propose the Quaternion-Hadamard Transformer Network (QHTN), a novel defense strategy against white-box adversarial attacks, including the Fast Gradient Sign Method (FGSM) and Projected Gradient Descent (PGD). The QHTN leverages a transformer architecture with three key modules: preprocessing, local-global feature extraction, and reconstruction. The local-global feature extraction module utilizes innovative Hadamard and quaternion convolution blocks to analyze spatial and inter-channel relationships. This unique approach enables the QHTN to incorporate a denoising mechanism during preprocessing, effectively mitigating adversarial noise before it influences the model's input. Extensive evaluations demonstrate the QHTN's efficacy in safeguarding haze and rain removal models from adversarial attacks. These results validate the QHTN's efficiency and potential for broader adoption in image-processing defense mechanisms.

(2024) Published by SPIE. Downloading of the abstract is permitted for personal use only.

Citation Download Citation

Vladimir Frants and Sos Agaian "Enhancing robustness of weather removal: preprocessing-based defense against adversarial attacks", Proc. SPIE 13033, Multimodal Image Exploitation and Learning 2024, 130330L (7 June 2024); https://doi.org/10.1117/12.3019864

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available