Learning-to-augment strategy using noisy and denoised data: Improving generalizability of deep CNN for the detection of COVID-19 in X-ray images

Mohammad Momeny, Ali Asghar Neshat, Mohammad Arafat Hussain, Solmaz Kia, Mahmoud Marhamati, Ahmad Jahanbakhshi, Ghassan Hamarneh

Abstract: Highlights•We introduce (i) noising- and (ii) denoising-based data augmenters to improve the generalization of a deep CNN.•We propose a “learning-to-augment” strategy to generate new data, noisy and denoised images.•Our approach outperforms state-of-the-art data augmentation methods when applied to COVID-19 detection on chest X-ray.