Author Topic: Registered Image Anomaly  (Read 689 times)

Offline vitozilla

  • Newcomer
  • Posts: 9
    • View Profile
Registered Image Anomaly
« on: 2019 August 08 19:52:04 »
I use a KAF3200 chip, and blooms are not new to me, but the recent PI update seems to have done something to the way registered images are pprocessed.  After images are calibrated, I run star alignment.  What I'm finding are the images get aligned, but the individual images have been modified, such that when integrated, blooms are out of control.  I must be missing something, but any clarification would be appreciated.  I have attached a snip that shows from left to right, an individual calibrated frame, it's subsequent alignment frame, then the integration of 15 aligned subs.

Offline Juan Conejero

  • PTeam Member
  • PixInsight Jedi Grand Master
  • ********
  • Posts: 7096
    • View Profile
    • http://pixinsight.com/
Re: Registered Image Anomaly
« Reply #1 on: 2019 August 09 03:58:24 »
I cannot reproduce this. Can you please upload a set of frames where this problem can be reproduced?
Juan Conejero
PixInsight Development Team
http://pixinsight.com/

Offline pfile

  • PTeam Member
  • PixInsight Jedi Grand Master
  • ********
  • Posts: 4691
    • View Profile
Re: Registered Image Anomaly
« Reply #2 on: 2019 August 09 12:08:46 »
wouldn't this just be because of your dither?

rob

Offline vitozilla

  • Newcomer
  • Posts: 9
    • View Profile
Re: Registered Image Anomaly
« Reply #3 on: 2019 August 09 20:53:22 »
Rob,

Dithering I think would help to get rid of outliers in a combined image, but why in a single registered frame would you have artifacts added from other frames in a register stack?  Shouldn't it just be aligning and rejecting outliers?  Like I said, I may be missing something, but I have not had this problem until recently.

Juan,

I sent you a private message, just let me know.

Offline pfile

  • PTeam Member
  • PixInsight Jedi Grand Master
  • ********
  • Posts: 4691
    • View Profile
Re: Registered Image Anomaly
« Reply #4 on: 2019 August 10 14:05:53 »
I'm just thinking that the dither puts the bright star in-between different pixels, thus making them slightly different from one another in each subframe... and then when you register and integrate, the blooms grow a bit since you're seeing the union of all the slightly different blooms.

rob

Offline vitozilla

  • Newcomer
  • Posts: 9
    • View Profile
Re: Registered Image Anomaly
« Reply #5 on: 2019 August 11 07:41:51 »
Rob,  Dithering puts everything in the image on a different pixel so that hot/cold pixels or any other chip imperfections (which will always be at the same spot, no matter whats in the image) never line up in a registered image, and therefore will be statistically rejected.  Blooms center on the bright stars that create them in the individual calibrated frames, and as such, if anything, they should also align with stars.  In the individual registered images, and as a consequence the master combined image, the width of the blooms has changed.  It appears the registering process does not move the blooms with the stars, and is averaging them out in individual registered images.  In my previous experience when registering calibrated frames, the blooms line up with the stars.  So my question is, in the latest update did something change about rejection parameters that I may have missed, or are there any other settings I may just be missing?

Offline pfile

  • PTeam Member
  • PixInsight Jedi Grand Master
  • ********
  • Posts: 4691
    • View Profile
Re: Registered Image Anomaly
« Reply #6 on: 2019 August 11 11:23:44 »
I know what dithering does. it also moves your bright star off of one set of pixels and onto another set of pixels, which might react differently to the charge overflow, thus changing the exact nature and width of the bloom.

just blink your registered frames and see if you see the blooms jumping around a tad. if not, my theory is wrong. if so, my theory may be right.

nothing has changed in ImageIntegration...

rob

Offline vitozilla

  • Newcomer
  • Posts: 9
    • View Profile
Re: Registered Image Anomaly
« Reply #7 on: 2019 August 11 15:25:42 »
Rob, I'm not trying to imply you don't know what dithering means.  I need help understanding that if no single calibrated sub has a bloom as wide as what is showing up in the registered subs, where is it coming from?  The blooms shouldn't be getting wider in registered images, correct?

You are kind of making my point though, that if the blooms are jumping around, wouldn't that make them statistical outliers, and as such get rejected?

I will send you a link to all my calibrated frames, take a look and see if I'm missing something, or you can repeat my results.  Thanks for your patience!

Kevin

Offline vitozilla

  • Newcomer
  • Posts: 9
    • View Profile
Re: Registered Image Anomaly
« Reply #8 on: 2019 August 17 07:21:01 »
Juan or Rob,

Have either one of you had an opportunity to access the link to the subframes?

Offline Juan Conejero

  • PTeam Member
  • PixInsight Jedi Grand Master
  • ********
  • Posts: 7096
    • View Profile
    • http://pixinsight.com/
Re: Registered Image Anomaly
« Reply #9 on: 2019 August 29 05:39:38 »
Hi Kevin,

I apologize for the delay in getting back to you. Thank you for uploading the images. This is an interpolation ringing problem with the default Lanczos interpolation used by StarAlignment. The bloomings are jump discontinuities that tend to generate strong ringing artifacts with any interpolation using an interpolation kernel with high-pass components. The ringing clamping algorithm prevents dark ringing, but at the cost of generating bright artifacts around the bloomings. This is a difficult problem but fortunately we have a good solution: drizzle.

In StarAlignment, use bicubic spline interpolation. This will still generate some ringing artifacts, but much smaller than Lanczos. Another good option in this particular case (not on a general basis!) is bilinear interpolation, which cannot generate ringing because it does not apply any high-pass filtering. The price to pay is much more aliasing, which reduces resolution. But this is not a practical problem if you use drizzle at the end of the process, since the registered images won't be used to generate the final integrated image.

Remember to enable generation of drizzle files in StarAlignment. Then use ImageIntegration to update the same drizzle files with pixel rejection and statistical data. Finally, use DrizzleIntegration to generate your final integrated image. Set drizzle scale = 1 if you want to preserve the original image scale and get better SNR, or try with drizzle scale = 2 if you prefer to work with a better modeled PSF at the cost or more noise. You have a good bunch of images, so you can try several options.

The *huge* benefit of drizzle integration is the total absence of pixel interpolation. This means absolutely no ringing and no aliasing artifacts, which is really great. In my opinion, DrizzleIntegration is the best option, in general, to achieve optimal results with any type of raw data in PixInsight. If using raw CFA data, enable CFA drizzle. If the images are undersampled and the amount of frames is reasonable, use drizzle x2. Otherwise, or if preserving the original scale is preferred, use drizzle x1.
Juan Conejero
PixInsight Development Team
http://pixinsight.com/

Offline pfile

  • PTeam Member
  • PixInsight Jedi Grand Master
  • ********
  • Posts: 4691
    • View Profile
Re: Registered Image Anomaly
« Reply #10 on: 2019 August 29 07:22:26 »
regardless the blooms themselves grow and shrink with the dither as i suspected - this can be seen in the unregistered images.

rob

Offline vitozilla

  • Newcomer
  • Posts: 9
    • View Profile
Re: Registered Image Anomaly
« Reply #11 on: 2019 August 29 19:31:07 »
Juan,  I greatly appreciate your patience and instruction.  The process you discuss was able to generate a much improved registration of images while keeping the blooms from getting out of control.  Needless to say, the issue is not with PIX!

Thanks again,
Kevin

Offline Geoff

  • PixInsight Padawan
  • ****
  • Posts: 903
    • View Profile
Re: Registered Image Anomaly
« Reply #12 on: 2019 August 30 01:15:21 »
The question I have is why has this only just come to light? Surely people have been registering bloomed subs for ages and the problem does not seem to have arisen until now.
Geoff
Don't panic! (Douglas Adams)
Astrobin page at http://www.astrobin.com/users/Geoff/
Webpage (under construction) http://geoffsastro.smugmug.com/

Offline vitozilla

  • Newcomer
  • Posts: 9
    • View Profile
Re: Registered Image Anomaly
« Reply #13 on: 2019 August 30 05:26:52 »
Geoff, that was the fundamental question I had.  And to your point, I've never used drizzle before this.  Why this particular data set required it is a mystery, but the registration and integration steps Juan posited works.