Ground Truth Free Denoising by Optimal Transport

TL;DR

This paper introduces an unsupervised denoising approach using optimal transport and Wasserstein GANs, capable of handling various data types without requiring paired noisy-clean samples, relying only on noisy data and noise examples.

Contribution

It proposes a novel unsupervised denoising method based on optimal transport and Wasserstein GANs that does not need paired data, extending applicability to diverse data types.

Findings

Effective on images and signals

Does not require paired training data

Works with arbitrary noise types

Abstract

We present a learned unsupervised denoising method for arbitrary types of data, which we explore on images and one-dimensional signals. The training is solely based on samples of noisy data and examples of noise, which -- critically -- do not need to come in pairs. We only need the assumption that the noise is independent and additive (although we describe how this can be extended). The method rests on a Wasserstein Generative Adversarial Network setting, which utilizes two critics and one generator.