Noise level and SNR estimation in images with signal independent Gaussian noise

Join the conversation on this Module


We have released our new Module “Noise level and SNR estimation in singe images with signal independent Gaussian noise”

What it does:

It estimates the amount of noise and the signal-to-noise-ratio from a single image. You don’t need another noise free ground truth image.

Why you should give it a try:

Usually in your daily microscopy work, you don’t have access to calibrated ground truth images for your samples to calculate the noise. Knowing the noise contribution in your images can be important for later image pre-processing and analysis. In addition, for many deconvolution solutions you normally need at least a rough noise estimate.

Where is the catch?

There is no real catch. You should just consider the outputted values as what they are – estimates. The values can vary depending on the parameter settings and the image structure. Our tests with real ground truth images show that the output values usually come quite close to the expected values.

Where it will work:

For now, this only works for images with a dominating Gaussian noise component: Widefield microscopy, light sheet microscopy, photography etc.

Where it won’t work:

Images from microscopy systems with signal dependent Poisson/Shot noise. This noise type needs a different model for estimation, which is not yet implement in our Module. Typical systems with this noise type: Confocal, STED

I hope you’ll like it! Tell us what you want to see next!


From Team APEER