NSG vs Local Normalization

Luca Marinelli

Hi Adam,

I just watched the NSG script videos and I plan to start testing this technique on my data. I am excited, as you are, because of the photometric approach to data normalization. As for the gradient subtraction, however, I was wondering if you had compared Local Normalization to the NSG. In principle, both methods try to solve the same problem: large-scale inhomogeneity of the sky background and how it affects proper data normalization. In the NSG case, the inhomogeneity is removed (mostly) and then data is photometrically normalized, In the LN case, normalization is evaluated on a patch basis. It obviously still suffers from the noise bias that you mention in the videos but I was curious to hear your thoughts.

Thank you!

Luca

BlockHead

I have never been able to get LN to really work well... it produces artifact (darkness around bright features and things) and you cannot control the result- all you have is a 128-256 parameter thing. So far there are no artifacts with NSG. Keep in mind the subtle difference. With LN, the recommendation is to make a *gradient free* reference. That is NOT the case for NSG. We are just picking a frame with the minimum (best, simplest) gradient. This means the reference frame is not trying to act as a gradient reducing tool- though it may have a simpler gradient that the target frames will be matched to.

Finally, I like NSG in that 

1. You get to see the actual Normalized frames finally! I ask Juan YEARS AGO to be able to output the real normalized frames just in general. 

2. With NSG there isn't a separate intermediate step of generating *.xmnl files... which is a somewhat mysterious thing.

-the Blockhead

Luca Marinelli

Thanks Adam. All great points.