Noise2Void equation E[ni] = 0


I have a question about the Noise2Void paper.

Equation (4) E[ni] = 0 embodies the assumption that the distribution of the noise is zero-mean.

But as far as I know, for a CCD camera based image (or a raster-scanned confocal image) noise represents added signal, so it cannot have a mean of zero with respect to the positive non-zero real pixel value. Is this correct ?

Thanks a lot for the clarification and apologies if I misunderstood the equation

Hi @LPUoO,

The dominant source of noise in microscopy typically is shot-noise. Shot noise is caused by photons arriving at the camera chip in irregular moments (the arrival times are poisson distributed). This distribution is not symmetric, but zero-mean.

Another noise source is readout noise and is typically modeled as gaussian noise, which is also zero mean.

In a nutshell: if I give you every day some coins (on top of your salary) you might still End up without any extra money if on some days I give you a negative number if coins. You see?

Hope this helps,