Citation

Su, Dennis Chuan Seng and Sim, Kok Swee and Abas, Fazly Salleh (2025) Evaluation of Adversarial Noise Attacks on Ai Systems Using Image Processing Techniques. In: 15th IEEE Symposium on Computer Applications and Industrial Electronics, ISCAIE 2025, 24 May 2025 - 25 May 2025, Penang, Malaysia.

Text 83.pdf - Published Version Restricted to Repository staff only Download (528kB)

Abstract

The resilience and dependability of AI systems have been seriously threatened by adversarial noise attacks, especially in image processing applications where accuracy and precision are crucial. This work examines the processes, effects, and efficacy of adversarial noise attacks across different AI models concentrating on the Fast Gradient Sign Method (FGSM) attack technique with different noise intensities of epsilon values between 0.1 and 0.5. We investigate how adversarial perturbations impair decision-making and model performance and evaluate five defense mechanisms which are Adversarial Training, Input Transformation, Defense Distillation, Gradient Masking and Randomization. The results reveal that adversarial training constantly improves resilience, achieving 100% accuracy under lower ε values. However, there is a trade-off between robustness and accuracy on clean dataset, and the computational cost of some defenses may limit their application in real-time system. This study highlights the important research gaps, such as the need for adaptive and hybrid strategies to enhance model resistance to adversarial attacks and suggests future researc

Item Type:	Conference or Workshop Item (Paper)
Uncontrolled Keywords:	Noise attack (FGSM), adversarial training
Subjects:	T Technology > TA Engineering (General). Civil engineering (General) > TA1501-1820 Applied optics. Photonics
Divisions:	Faculty of Engineering and Technology (FET)
Depositing User:	Ms Rosnani Abd Wahab
Date Deposited:	18 Mar 2026 08:01
Last Modified:	19 Mar 2026 00:48
URII:	http://shdl.mmu.edu.my/id/eprint/15569

Downloads

Edit (login required)