Denoising structured noises without clean data (Convallaria)
Authorship
The data was acquired by the LMF at Max Planck Institute for Cell Biology and Genetics and originats from this publication: Removing Structured Noise with Self-Supervised Blind-Spot Networks by Coleman Broaddus et. al.
Details
Self-supervised denoising methods may not work properly if the noise is not pixel independent and presents patterns or structures in the image. If this is the case, some methods may augment the artifacts and removing them may become complicated. Struct Noise2Void is an extension of Noise2Void that tackles the problem of denoising images which have structured and pixel-dependent noise patterns.
This example scenario shows a 2D image from Convallaria majalis acquired with a Spinning Disk confocal microscope, where horizontal stripes of noise can be observed.
Demo
A practical notebook example with python can be found here.