›

Pixinsight recommends always using Drizzle, even at scale 1? An experiment but still unclear to me

Reinhard Gloggengiesser
edited September 2023 in PixInsight

I just came across a statement from the official Pixinsight YouTube channel that surprised me. In this video https://www.youtube....h?v=7OZzfmCYmeA at 6:40 when introducing the newest version of WBPP the narrator says that they "recommend always using Drizzle" and then proceeds to use the scale of 1, meaning that the final image resolution doesn't get increased. I usually never use Drizzle because with my rig I'm always oversampled, but this made me think. I wonder what's behind that reasoning? Does anybody have an idea? 

 

To try it out, I ran 45 light frames (plus all calibration frames) of the Cave Nebula through the complete WBPP process with Drizzling enabled. The good thing is that WBPP then outputs both a non-drizzled and a drizzled master light, so it's easy to compare the results.

 

I attach 2 frames with just a simple STF "nuclear" stretch, drizzled x 1 on the left and non-drizzled on the right.


image



What is immediately noticeable is that the drizzled master seems to show more contrast, sitting about halfway between the non-drizzled master and the non-drizzled master binned by 2 (not shown). So by simply drizzling by a scale factor of 1, do I get better signal to noise ratio, but not the downside of loss of resolution I would get by binning? The noise also seems to go up slightly, but that seems manageable in the days of NoiseXTerminator etc.


When checking for the SNR of the two images, SubFrameSelector gives me the following result:


image


 

So what's going on there? Does drizzling result in increasing SNR? And if so, how? I couldn't find an explanation anywhere, but maybe I'm just too clumsy to find it.


P.S. If needed, I can upload the original master lights

Comments

  • BlockHeadBlockHead
    You are correct. 
    And I do not know.
    The original idea was the 1x1 with CFA Drizzle had benefits for avoiding issues with color calibration due to the vagaries of interpolation. 

    I have asked for clarification on the official forum. 
    I think it is an error...a surprising one at that.
    If it is NOT an error... it is said completely without explanation which is a WORSE issue.

    -the Blockhead



  • BlockHeadBlockHead
    The reasoning is to avoid artifacts of resampling that usually happens with registration of images. 
    So...yes, the advice is to do...and I am happy to do so and go around and tell people to do so as well..but I want to understand the limits of this advice.

    See:

    -the Blockhead
  • Reinhard Gloggengiesser
    Thanks Adam for your feedback and for getting on this in the PI forum! I am very much intrigued by the responses there and will definitely continue watching the thread you dug up.
  • Andrea GironesAndrea Girones
    HI Adam and Reinhard i was looking into this issue as well as I have a huge number of OSC subs that are undersampled. I am keen to drizzle this data and i have been reviewing the WBPP classes  mainly to  remmeber how to use subframe weighting and local normalization because I have so many images over many nights.

    But then  i noticed the drizzle section in WBPP  had a 1x option which stumped me. Clearly your videos recommend 2x drizzle which does of course make a huge image.
    If I could get the same drizzle benefit using 1x drizzle I would like that . 

    My other option is to drizzle 2x.  I considered whether to then then downsample back to 1x to save drive space and image processing time mainly. Or is that a silly idea because I have then lost the benefit of drizzling?
    What advice would you have?  
    Thanks

    Andrea
  • BlockHeadBlockHead

    Assuming you have well dithered data and 15 or more frames... 

    Drizzle has two features.

    1. Avoiding the pitfalls of interpolation artifacts that come with regular StarAlignment (registration). (Drizzle 1x1)
    2. Improving the quality of spatial information in undersampled images. Drizzle 2x2

    It appears that #1 is always beneficial and is particularly necessary for OSC data. #2 is optional for undersampled data- but can be helpful if there is very bright non-stellar information you would like BXT to improve. 

    #1 is included in #2.

    Does this answer the question?
    -the Blockhead
  • Andrea GironesAndrea Girones
     Ok got it!   So 1x is recommend for all OSC images and 2x may  help under sampled date like mine…. Excellent thanks!
    Back to 2x drizzling to fix the blocky stars !!!

    What you said about BX was interesting… my data has a very faint object…. Which is why I gathered so much data …
    Andrea
  • Dr.LesDr.Les
    Do I understand correctly, that the 1x Drizzle does NOT increase the file size ?

    Thx.

    L
  • BlockHeadBlockHead
    Correct. It is just the assembly (integration) of the images does not use regular interpolation.
    -the Blockhead
  • Dr.LesDr.Les
    After I read your comment, I looked at my most recent masters from a project.  The drizzle file is about half the size of the undrizzled flile.  See bottom two files in screen shot.  Also attaching the WBPP pipeline, if it is useful.

    I also do not understand why the size of image files change if they have the same resolution.  Does it take more or less space depending how bright pixels are ?

    Thanks,  Les
    Master files Screenshot 2023-10-17 105716.jpg
    804 x 287 - 77K
    wbpp success Screenshot 2023-10-17 105211.jpg
    1061 x 952 - 355K
  • BlockHeadBlockHead
    edited October 2023
    No... you should open them and confirm that size of the images are the same number of pixels.
    I suspect the difference is that the other masters have the rejection maps included (you will see them when you open). This certainly bloats the memory storage.

    -the Blockhead
Sign In or Register to comment.