Adaptive Normalisation : bug ?

Nov 20, 2020
5
0
Hi,

My first post on this interesting forum. First of all thanks for the sharing of information that can be found here.

My problem is as follows :
- I took 80 photo's with a Nikon Z7 and 300mm f4 PF lens and ioptron skyguider pro.
- image calibration, debayer, subframeselector with the weights of lightvortex website, staralign, and then imageintegration.
- Normalisation : Average, Normalisation : Adaptive normalisation, FITS keyword SSWEIGHT.
- Pixel Rejection : Linear fit (or winsorized sigma clipping) and adaptive normalisation.
- in fact, the first time I did the stacking, I had a star align with another photo from another session that was turned 90 degrees. The colored zones were not present, but there was another less obious blue zone. I thought a reflection, but that is not possible as on this photo it is not present. So the artefact changes in function of the content of the photo. That was with a previous version of PI.

- The two zones are also visible in the high and low rejection photo's.

- If I do not use Adaptive Normalisation, the artefacts are not present.

-Afterwards I simply stretched with an STF and then a HT with the parameters from the STF. No other processes were done on this picture. Resized for publication in photoshop.

Windows 10
PixInsight version : 1.8.8-7

Seems something goes wrong with the algorithm ?

The blobs seem analogous to what can be seen in this older posting :
 

Attachments

Nov 20, 2020
5
0
I just read this posting, where Juan explains why that normalisation is not necessary in this case. The danger of using a technique one does not really understand...

But, it is by making errors that one learns. I've learned something...

 

pfile

PTeam Member
Nov 23, 2009
6,170
180
god bless light vortex but the inclusion of adaptivenormalization in that tutorial has caused a lot of electrons to be spilled here. unless you have super-bad LP it's probably not necessary. and if you are going to use it you need to come up with a very clean reference image. i'd say in most cases it's something to experiment with but it should not be a default step in your flow.

rob
 

Linwood

Well-known member
Jul 28, 2020
125
9
if you are going to use it you need to come up with a very clean reference image. i'd say in most cases it's something to experiment with but it should not be a default step in your flow.
Can you comment a bit on what "clean" here means?

For example, for alignment I usually pick 5 or so very good (low FWHM and eccentricity) subs and align and integrate to an alignment master.

But I have never know really what to do about integration reference -- is it best to pick one with the most even background and not worry much about star size or shape? Or do we also look here for lowest FWHM/eccentricity?

Should one actually run DBE on the reference image first perhaps? Or otherwise 'clean' it?

In other words -- can you define clean?
 

pfile

PTeam Member
Nov 23, 2009
6,170
180
the LocalNormalization reference needs to be gradient free, or localnormalization will insert any gradients in the reference into the subs... and you don't want that. so yeah, DBE would be a good idea. but of course if you are not careful with DBE you might introduce artifacts which will then be picked up by LN.

rob
 

pfile

PTeam Member
Nov 23, 2009
6,170
180
i actually didn't notice that the OP was talking about AN in the first post and then LN in the 2nd (despite the thread title)! i can't find anything in the documentation that would indicate that the reference image needs to be particularly 'clean' for AN to work.

rob
 
  • Like
Reactions: Linwood