University of Twente Student Theses


Data Augmentation using Fourier-Basis Noise

Fattahi Mehr, Omid (2022) Data Augmentation using Fourier-Basis Noise.

[img] PDF
Abstract:Data augmentation has become an important tool to improve the robustness of a model against corruptions in data and adversarial attacks. Most of the previous research has focused on approaching data augmentation from the spatial domain. This paper utilizes Fourier-Basis noise to augment images. Fourier-basis noise consists of frequencies added to an image. We define a new selection method by creating predefined frequency sets on different criteria. These sets are simpler than other established methods and have many possible configurations in which different frequencies can be combined. We conduct experiments that are used to evaluate the effect of those sets on the robustness against common corruptions. The results show that high frequency noise augmentation provides a significant improvement in robustness against corruptions compared to the baseline model. This research shows positive results on the effect, Fourier-Basis noise can have on corruption robustness and suggests further exploration of the method for a better understanding of its impact.
Item Type:Essay (Bachelor)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Computer Science BSc (56964)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page