Fully Unsupervised Probabilistic Noise2Void

TL;DR

This paper advances unsupervised deep learning for biomedical image denoising by making Probabilistic Noise2Void fully unsupervised through parametric noise models that do not require calibration data.

Contribution

It introduces improvements to PN2V, replacing histogram models with parametric ones and enabling noise model creation without calibration data, making PN2V fully unsupervised.

Findings

Parametric noise models outperform histogram-based models.

The method achieves comparable denoising quality without calibration data.

All proposed improvements are practically effective.

Abstract

Image denoising is the first step in many biomedical image analysis pipelines and Deep Learning (DL) based methods are currently best performing. A new category of DL methods such as Noise2Void or Noise2Self can be used fully unsupervised, requiring nothing but the noisy data. However, this comes at the price of reduced reconstruction quality. The recently proposed Probabilistic Noise2Void (PN2V) improves results, but requires an additional noise model for which calibration data needs to be acquired. Here, we present improvements to PN2V that (i) replace histogram based noise models by parametric noise models, and (ii) show how suitable noise models can be created even in the absence of calibration data. This is a major step since it actually renders PN2V fully unsupervised. We demonstrate that all proposed improvements are not only academic but indeed relevant.