Preprocess Normalization

[Morten Lerager]

Hi'
I did som test with Normalization in preprocess options Tab.
Test done with OSC Camera
Without any normalization I get :
Sigma: Rejected pixels = 1.7 %
Calculating SNR
SNR = 49.05 db, R Preprocessing_Without
SNR = 46.47 db, G Preprocessing_Without
SNR = 48.89 db, B Preprocessing_Without

With normalization set to ADD_Sigma_3
Sigma: Rejected pixels = 1.4 %
Calculating SNR
SNR = 50.82 db, R Preprocessing_ADD_Sigma_3
SNR = 48.25 db, G Preprocessing_ADD_Sigma_3
SNR = 50.71 db, B Preprocessing_ADD_Sigma_3

With normalization set to ADD Sigma 8
Sigma: Rejected pixels = 0.1 %
Calculating SNR
SNR = 52.38 db, R Preprocessing_ADD_Sigma_8
SNR = 48.59 db, G Preprocessing_ADD_Sigma_8
SNR = 52.21 db, B Preprocessing_ADD_Sigma_8

With normalization sigma high gives a SNR a little better.
But the higher I set Sigma gives less rejected pixels. ( isn't that bad )
How can i decide which are the best ?

[fabdev]

Hi, when the Sigma coefficient goes up and up, the Sigma algorithm becomes a classical Average/Add (without Sigma), which gives the best SNR.

Sigma is useful to remove hot pixels and satellite stripes, but if these are not present, then Average/Add is better (also Sigma cannot be used for precise astrometry and photometry). So, I usually start with Average/Add and switch to Sigma only if needed.

Sigma coefficient between 3.0 and 5.0 are good for most purposes. Coefficient 8.0 may be very similar to simple Add/Average.
For extremely noisy images 1.5 - 2.0 may be necessary.
So, the best Sigma is the higher one which is able to delete hot pixels and satellite stripes. Sigma infinite = simple Average/Add

O.T. In physics, 5.0 is often used:
https://home.cern/resources/faqs/five-sigma

[Morten Lerager]

Thank you Fabio.
At least I now know what it' does.
So I will continue be using Sigma+Sum for Light stacking.
In preprocess options > normalization > Images ADD Sigma 8
And take sigma down if there still are sattelite trails.