›

Newest LocalNormalization vs NSG?

James Sokolowski
in PixInsight
NSG seems to be a very good tool for preparing weights for subs for ImageIntegration and I've been using it 'successfully' (I think !!??) for a while now.  W/ the latest PixInsight we have an improved LocalNormalization and I'm wondering if folks are getting a sesne that this is the direction PI is going to either incorporate the NSG algorithms more fully into PI or offer a different but competing capability??

Happy New Year !!

Jim

Comments

  • BlockHeadBlockHead
    At the moment Juan is moving in the direction of a different algorithmic way to both normalize and determine weights. So NSG will remain one kind of method. 

    The behind-the-scenes info is that Juan algorithms are beginning to converge/agree with NSG- but due to the way the measurements are made- there may *not* be agreement in certain conditions (clouds, poor seeing...etc). This is the sticking point actually. So it is still a work in progress. One of the difficulties is that Juan believes that the photometry done in NSG is not necessarily the only way to arrive at good answers- which is fine- but he would go on to say that his methods stand on equal footing. This means you can't use NSG to serve as a comparative test of results. Unfortunately this makes things difficult to assess unless you dig deep into the math.

    My opinion is that whatever method is chosen- there are real-world observables that I think everyone can agree on. For example, extincted images observed at high air-mass should have lower weights than low-airmass observations of the same field. The photometry that these methods use should be in agreement on this fact. There are many kinds of observables. The brightness of the background should not affect the flux you measure... all kinds of stuff. 

    So, the bottom line is the Juan recognizes that this area of image analysis need improvement and he is really really working on it...but it is a work in progress. 

    -the Blockhead
  • James Sokolowski
    Took the M 51 data from your website (via fundamentals / workflows) and ran both LN and NSG w/ same ref frame on L, R, G, B, & Ha... as best I could set things up.  I checked the two images for each filter against one another in SubframeSelector and got very similar results for measurables PSFs (both), FWHM, E, stars, SNR, etc for all bands except Red.  In Red, the LN result showed curious dark rings around the brightest half of the stars... However, I didn't go back and double check all my setups for that filter or try to determine what might have caused that flaw

    For now, NSG remains my preferred option...  I like the plethora of settings and wetup checks one can use and it doesn't hurt that it sets up ImageIntegration for me too !!

    Jim
  • BlockHeadBlockHead
    The dark ringing is a indeed an artifact of LN. This is the issue with that method for many data.
    Also, using the standard method, Weighting and Normalization are handled as different uncorrelated things. NSG correlates the weights with the measured flux (which is my preference).

    The PSF Signal that Juan is working on... even in this version, is still a work in progress.
    -the Blockhead
  • Marsha Ann Wilcox
    Thanks for this discussion, very helpful. 

    I updated PI last night and installed NSG as a separate script and read what I could find about the change. I guessed that if anyone had compared the output from the new LN and NSG, I’d find it here.  Seems it’s not time to leave NSG just yet - nice to have 2 tools.

    Thanks again,
    Marsha
  • James Sokolowski
    I've been working on 5 nights of M 101 data, where I used NSG before the new PI release.  W/ some weird family health issues I've been distracted so screwed up the flow and will have to go back, but I have two new nice nights on M 106 and when I process, I'm going to try to make an objective-ish comparison of the two methods.

    I like the tech aspects of the new Signal / Noise / SNR measures in PI... makes sense (excpet I didn't understand why a fit to the star profiles would be made, but then the photometry would use the original data instead of an integration over that fit... normally one of the best reasons to do any kind of fit is you get a better approximation of the underlying distribution... so would seem using the fit to do the photometry would have been a more robust move...

    I hope to move onto the M 106 data in the next week or so and look forward to working both LN and NSG on it... but of course, w/ the other new tools, there seem many ways to work the new LN workflow, w/ various image weightings based on the new script... forget its name...  but I tried it out and it works nicely...

    Jim
  • James Sokolowski
    Just did another "test" of NSG and the newest LN.  Ran both on the NGC 6888 data available through Adam Block Studios Horizons collection.  Don't remember all the settings for either... sorry, could look them up if anybody really wants... I saved the R, G, B, L, Ha results for the two, then averaged the two Ha's in PixelMath to form a "synthetic luminance" as both have great spatial detail... OK... I'm attaching an image of the results of SubFrameSelector for PSF Signal Weight and PSF SNR... both show the LN results consistently better than the NSG (not saying there couldn't be operator error here...) and the SynLum is better in both than either of the two Ha frames individually (so that makes sense).

    Order of the files is LN first, then immediately after the NSG.  Filters are Blue, Green, Lum, Red, Ha, with the SynLum after those...

    While nothing definitive, I'd argue that I'm getting very comparable results w/ LN and NSG here.  I also looked at "Stars" in SFS and found overalll the NSG results do provide higher star counts...

    Jim
    SFS LN vs NSG 03May22.png
    1555 x 1005 - 532K
  • James Sokolowski
    oh poop... attached the wrong plot... well the PSF SNR looks quite similar to PSF SW... "trust me" !!

    Jim
Sign In or Register to comment.