LocalNormalization producing weird stars

[joelshort]

I really want LN to work.  I can tell that it is a wonderful tool that greatly helps with background data.  For example, I'm currently working on an image of vdB158, a very faint wispy nebula surrounded by areas of dust.  With LN those areas of dust really stand out a lot more. 

Unfortunately I con't it to work well in two scenarios.  The first is with narrowband data.  The resulting stack comes out with too wide of a dynamic range.

The second scenario is what I'm more concerned about.  In my LRGB images LN produces wacky stars.  The outsides of the stars are not smooth.  Below is an example of what I mean. 

Does anyone have any suggestions for how to use LN but not produce this kind of artifact?

[pscammp]

Joe,

Try raising the 'Scale' figure, I was playing with this over the past week and if I dropped this value too much then
it began having strange effects on the larger stars etc

Of course I might just be talking a load of 'uneducated' rubbish but it's worth a go

Paul

[Juan Conejero]

Hi Joel,

I need a set of calibrated/registered images (the input of ImageIntegration) to help you. As I've said many times, LN is a powerful but complex and dangerous tool requiring a previous analysis of the data.

[joelshort]

Hi Joel,

I need a set of calibrated/registered images (the input of ImageIntegration) to help you. As I've said many times, LN is a powerful but complex and dangerous tool requiring a previous analysis of the data.

Thanks Juan.  I can send you some calibrated/registered files but the LUM data set consists of 200 images totaling 12GB.  Will a smaller set suffice, or is there a place I can upload to that can handle that amount of data transfer?

[joelshort]

I should mention that my solution to this was to simply produce an integrated image with and without LN and masked in the stars from the non-LN stack into the LN stacked image.  Worked great!

LN works extremely well.  I could not have pulled out the faint background dust in the image I am working on without it. 

[joelshort]

Just an update on this...I changed the LN scale to 256 instead of 128 and that seems to have remedied the star weirdness.  I will look forward to a tutorial when it becomes available.

[AstroScience]

Yes, Local Normalization definitely can kill the data or introduce data that is not "there" in reference image, we should be really careful using this tool while waiting for detailed explanation or tutorial.

As I understand it is completely depended on your imaging scale, as for what "scale" setting to put in LN.  For now I created different folders for each LN scale , like LN_32,LN_64 and so on.
Using process container I created LN instances for each scale and let the container run. Then again, using the process container I created each integration for each scale and let it run, saves a lot of time on testing.

After it finished I got 9 integrations , including one that I didn't used LN at all and compared the results. Scale of 32 pixels created worse artifacts, I simply tossed it. The next scales were showing definitely an improvement over a stack using no LN.
It was like regular stacking SNR improved by 2.28 and using LN the stack's SNR improved by 2.57. The next scales, up to 256 pixels showed a reduction in SNR, 2.49, 2.47, 2.46 etc.... but they all look the same...Can't really make up my mind on what scale to use to get better result.... really stumbled....

My gut feeling tells me to use the scale that produce the best SNR improvements in a stack while keeping it artifacts free, but how can we be sure that we not introduce a false data to a stack?