Journal article
Robust image classification using a Low-Pass activation function and DCT augmentation
IEEE Access, Vol.9, pp.86460-86474
2021
Abstract
Convolutional Neural Network’s (CNN’s) performance disparity on clean and corrupted datasets has recently come under scrutiny. In this work, we analyse common corruptions in the frequency domain, i.e., High Frequency corruptions (HFc, e.g., noise) and Low Frequency corruptions (LFc, e.g., blur). Although a simple solution to HFc is low-pass filtering, ReLU – a widely used Activation Function (AF), does not have any filtering mechanism. In this work, we instill low-pass filtering into the AF (LP-ReLU) to improve robustness against HFc. To deal with LFc, we complement LP-ReLU with Discrete Cosine Transform based augmentation. LP-ReLU, coupled with DCT augmentation, enables a deep network to tackle the entire spectrum of corruption. We use CIFAR-10-C and Tiny ImageNet-C for evaluation and demonstrate improvements of 5% and 7.3% in accuracy respectively, compared to the State-Of-The-Art (SOTA). We further evaluate our method’s stability on a variety of perturbations in CIFAR-10-P and Tiny ImageNet-P, achieving new SOTA in these experiments as well. To further strengthen our understanding regarding CNN’s lack of robustness, a decision space visualisation process is proposed and presented in this work.
Details
- Title
- Robust image classification using a Low-Pass activation function and DCT augmentation
- Authors/Creators
- M.T. Hossain (Author/Creator)S.W. Teng (Author/Creator) - Federation UniversityF. Sohel (Author/Creator) - Murdoch UniversityG. Lu (Author/Creator) - Federation University
- Publication Details
- IEEE Access, Vol.9, pp.86460-86474
- Publisher
- IEEE
- Identifiers
- 991005540116807891
- Copyright
- © 2021 IEEE
- Murdoch Affiliation
- School of Information Technology
- Language
- English
- Resource Type
- Journal article
UN Sustainable Development Goals (SDGs)
This output has contributed to the advancement of the following goals:
Source: InCites
Metrics
29 Record Views
InCites Highlights
These are selected metrics from InCites Benchmarking & Analytics tool, related to this output
- Collaboration types
- Domestic collaboration
- Citation topics
- 4 Electrical Engineering, Electronics & Computer Science
- 4.17 Computer Vision & Graphics
- 4.17.128 Deep Visual Recognition
- Web Of Science research areas
- Computer Science, Information Systems
- Engineering, Electrical & Electronic
- Telecommunications
- ESI research areas
- Engineering