Jump to content


Photo

Drizzle idiocy

  • Please log in to reply
15 replies to this topic

#1 bunyon

bunyon

    Gemini

  • *****
  • Posts: 3274
  • Joined: 23 Oct 2010
  • Loc: Winston-Salem, NC

Posted 20 February 2013 - 09:57 AM

I've commented several times that I've never gotten anything out of drizzle on AS!2. I made a "breakthrough" and was very proud of myself until I realized I had probably been doing things stupidly.

The breakthrough was to resize the drizzled stack BEFORE applying wavelets. I made this intellectual leap while reading another thread. It occurs to me that resizing before wavelets makes all kinds of sense, given what wavelets do.

So, is this common knowledge and I've simply been doing a silly thing? (I also post in case anyone else is missing this step).

For what it's worth, I got very nice 1.5x and 3x drizzle stacks from one of my nicer Jupiters. I can't say it's obviously better than without the drizzle, but at least it's not a fuzzy mess.

#2 MvZ

MvZ

    Surveyor 1

  • *****
  • Posts: 1580
  • Joined: 03 Apr 2007
  • Loc: The Netherlands

Posted 20 February 2013 - 10:27 AM

with resizing I'm assuming you mean 'making the 3x stack smaller'? That makes sense to get rid of grid artifacts, and that is exactly what the 1.5x drizzle stack actually is (a resized 3x drizzle stack).

#3 bunyon

bunyon

    Gemini

  • *****
  • Posts: 3274
  • Joined: 23 Oct 2010
  • Loc: Winston-Salem, NC

Posted 20 February 2013 - 10:54 AM

Right. Resized both drizzles back to 648x488, so that all the images taken to wavelets were the same size.

#4 tjensen

tjensen

    Mercury-Atlas

  • -----
  • Posts: 2609
  • Joined: 16 Feb 2005
  • Loc: Chapel Hill, NC

Posted 20 February 2013 - 11:52 AM

:foreheadslap:

#5 MvZ

MvZ

    Surveyor 1

  • *****
  • Posts: 1580
  • Joined: 03 Apr 2007
  • Loc: The Netherlands

Posted 20 February 2013 - 12:19 PM

Resized back to the original size??

Then you are basically not using drizzling for what the technique actually does: increasing resolution for undersampled images.

If resized to the original size there is no real image improvement, except for perhaps an unmeasurably small improvement of alignment accuracy. You might as well just go for the 9x faster normal processing.

Furthermore, the effects of the image resizing algorithm you used (yes, resizing messes with data!) is probably much bigger than what drizzling did to the image. Resizing usually either blurs or sharpens the image a bit, so there is no way to directly compare an original stack and a 'resized back to original size' drizzled stack, making it pretty much impossible to make any valid claims regarding 'drizzling'.

#6 bunyon

bunyon

    Gemini

  • *****
  • Posts: 3274
  • Joined: 23 Oct 2010
  • Loc: Winston-Salem, NC

Posted 20 February 2013 - 12:30 PM

Okay, so I wasn't off base originally.

In that case, drizzling has never yielded a usable image of any sort - it is fuzzy to the point of almost no detail and when any of the various sharpening techniques are applied, the image only gets worse (this includes wavelets in R6 and LR deconvolution in AstraImage).

In that case, can you think of what I'm doing "wrong"? Is it simply a case of the data isn't undersampled, so there should be no improvement with drizzling?

#7 MvZ

MvZ

    Surveyor 1

  • *****
  • Posts: 1580
  • Joined: 03 Apr 2007
  • Loc: The Netherlands

Posted 20 February 2013 - 01:10 PM

> Is it simply a case of the data isn't undersampled, so there should be no improvement with drizzling?

Likely, but that doesn't mean the image should look bad.

You MUST however adapt your further processing to the drizzling output (e.g. use coarser sharpening, deal with potential artifacts, etc etc..)

Perhaps it is a good idea to just show your images, otherwise it will be difficult to give pointers.

Either way, drizzling IS NOT just a magic make-my-image-look-better option. With careful processing it CAN make your image look better under certain conditions. With careful processing there is no reason it should make your images look worse.

#8 Kokatha man

Kokatha man

    Fly Me to the Moon

  • *****
  • Posts: 6941
  • Joined: 13 Sep 2009
  • Loc: "cooker-ta man" downunda...

Posted 20 February 2013 - 07:10 PM

.....just by way of a practical experience in the benefits of 3X drizzling I was very unconvinced of its merits untill fairly recently...

My "conversion" came by way of the ASI120MM where although I haven't done the maths I am almost sure I am under-sampling with all the images I have captured to date.....sure I am waiting for the opportunity to use a greater image scale but imo seeing has to co-operate as well for optimal outcomes regardless of the maths, and I just haven't experienced what I'd term "very good' seeing so far.....

So, if my assumption about under-sampling is correct then my findings would be a neat demonstration of what Emil stated above - that is, I find my ASI120MM images respond extremely well to 3X drizzling in AS!2.....with the (also) stated harder/coarser wavelets applications part of the pre-requisites!

...just my 2 cents worth! :penny: :penny: :grin:

#9 bunyon

bunyon

    Gemini

  • *****
  • Posts: 3274
  • Joined: 23 Oct 2010
  • Loc: Winston-Salem, NC

Posted 20 February 2013 - 08:36 PM

I did some math and I am, generally, oversampling. That may have something to do with it.

And, Emil, I apologize if it came off like I expected a magic bullet. I don't. Although, I must say, I find AS!2, in general, to be a magic bullet, so I can't really ask for two bullets!

My only concern is I hear tell of drizzling being a useful tool from some folks whereas it does either nothing or worse than nothing to my images. In such situations, I usually assume I am the problem. I'll dig up some attempts tomorrow and post.

#10 ToxMan

ToxMan

    Mercury-Atlas

  • *****
  • Posts: 2777
  • Joined: 23 Jan 2011
  • Loc: Tucson, Arizona, USA

Posted 21 February 2013 - 12:40 AM

Would over-sampling cause more artifacts, like Moire patterns? Seems like I had more trouble with this artifact when I tried to go to f25 with C8 and DMK31, trying to shoot Jupiter. I'd get a larger image on the chip. But, too dim and gain had to be set high. Or, at max. Naturally, making the problem worse in the red filter video.

#11 bunyon

bunyon

    Gemini

  • *****
  • Posts: 3274
  • Joined: 23 Oct 2010
  • Loc: Winston-Salem, NC

Posted 21 February 2013 - 06:12 AM

Not sure about artifacts but I can confirm that in an undersampled avi (see my other thread on the moon), that drizzle works fine and the final images are, perhaps, a little better than without. In any case, the drizzled images can be sharpened and manipulated in a way that my previous attempts can't.

So, at this point, I'm willing to conclude that my problem in other images is that they're oversampled to begin with.

#12 tjensen

tjensen

    Mercury-Atlas

  • -----
  • Posts: 2609
  • Joined: 16 Feb 2005
  • Loc: Chapel Hill, NC

Posted 21 February 2013 - 07:17 AM

Hey Paul,
Drizzling was designed for undersampled images and doesn't really work if they are over sampled (as stated above). Have you tried doing LR decon on your Nyquist/oversampled images and then reducing to native capture size? I've found that can work. Naturally, all of this is seeing dependent.
Cheers

#13 ToxMan

ToxMan

    Mercury-Atlas

  • *****
  • Posts: 2777
  • Joined: 23 Jan 2011
  • Loc: Tucson, Arizona, USA

Posted 21 February 2013 - 08:57 AM

Yes I have, Tim. It wasn't until I sorta understood this subject that I understood why I was having difficulties. My take-a-way, if the seeing is not so good, don't try to make such a large image on the chip. And, over-sampling can introduce some uncontrollable artifacts. If the seeing is good, up that image size. Use drizzle for under-sampled data.

So, Paul's point is, drizzle before wavelet sharpening. Emil's point is, don't take image back down to native resolution because there is little benefit from drizzle used in this manner. About right?

#14 bunyon

bunyon

    Gemini

  • *****
  • Posts: 3274
  • Joined: 23 Oct 2010
  • Loc: Winston-Salem, NC

Posted 21 February 2013 - 05:40 PM

Going back through my data, images that are undersampled do work out okay with drizzle. Oversampled don't. So there is that.

I put this up in the solar forum but I'll link it here, too. This is a mosaic of 9 panels shot undersampled through a PST. The images were stacked in AS!2 with 1.5x drizzle and the resulting images sharpened and behaved like an image should. So, I think I'm now down with drizzling.

Posted Image

#15 ToxMan

ToxMan

    Mercury-Atlas

  • *****
  • Posts: 2777
  • Joined: 23 Jan 2011
  • Loc: Tucson, Arizona, USA

Posted 21 February 2013 - 10:33 PM

This thread has me looking at some "old" under-sampled data, too. Thanks, gents.

#16 Mike Phillips

Mike Phillips

    Soyuz

  • *****
  • Posts: 3522
  • Joined: 21 Feb 2006
  • Loc: Swift Creek, NC - 35.682 N, 78.743°W

Posted 22 February 2013 - 11:48 AM

This thread cuts right to the heart of imaging now doesn't it. Oversampling and understamping are terms I have a vague idea about. In most cases it's something something most people think about or have much control over.

In terms of processing I've always done a 1.5x upscale or even 3x for Mars either using ninox or AS2. I never really thought much of drizzle/resize and it's relation to image scale / focal-ratio.

For certain in poor seeing the same optical train and processing gives me all sorts of fits for artifacts etc, especially on Saturn at full gain!

I'm favorit'ing this thread to come back to later for sure, nice job Paul's, Emil, Tim and Darryl!

Mike






Cloudy Nights LLC
Cloudy Nights Sponsor: Astronomics