Fast Fourier Convolutions in Self-Supervised Neural Networks for Image Denoising

Joonas Ariva, Mikhail Papkov

Keywords: Fast Fourier Convolution, denoising, self-supervised learning, Noise2Same, computer vision

TL;DR: We design multiple ways for imputing Fast Fourier Convolutions into self-supervised neural networks for image denoising.

Abstract: Recently, denoising convolutional neural networks (CNN) have started to outperform classical denoising algorithms. However, CNNs performance could be constrained by the limited receptive field of regular convolution. To mitigate this problem, a new modification for CNNs was proposed: Fast Fourier Convolution (FFC). Here, a global receptive field is achieved by using Fourier Transform and convolving spectral representation. The global perception field can help CNNs to better capture dependencies in image regions that are far apart. In this work, we design multiple approaches for incorporating FFC into self-supervised neural networks for image denoising. We evaluate these approaches on three benchmark datasets and compare them with supervised and self-supervised methods. We empirically show that an FFC-enhanced denoising network achieves the state-of-the- art results on the character dataset and shows a comparable level of performance for both grayscale and color natural images.