Drizzling 1x vs 2x--So That's How the Myth That Drizzling Degrades Well-Sampled Images Happened
ฝัง
- เผยแพร่เมื่อ 7 พ.ค. 2024
- Let's explore an assumption and a question that were brought up after my last two videos on the advantages of drizzling. The assumption was: Drizzling well sampled information is pointless if the information came from a telescope of greater than 1000 mm focal length. How accurate did actual experimentation prove this statement to be? And the question was: If your image is well sampled, isn't it better to drizzle 1x instead of 2x? Experimentation shows the answer is a little complex, and even revealed how the myth that drizzling is bad for well sampled images may have evolved. But experimentation also revealed there is a way to take advantage of the outcomes and make 1x and 2x drizzling work together to improve your images even more.
You can view the fully developed drizzled 2x image of the Ghost Nebula here: www.astrobin.com/xorr7d/
You can view higher res versions of all five of the images discussed in this video here and draw your own informed conclusions: astrob.in/wzvztq/E/
An abridged, semi-technical paper on the development of drizzling for the Hubble Space Telescope. www.stsci.edu/~fruchter/dithe...
*** *** ***
Check out a portfolio of astro imagery shot from the Sky Story Observatory at Astrobin: www.astrobin.com/users/SkyStory/
We are Telescope Canada's first affiliate. It's a great place for astrophotography gear. Using the following link adds nothing to your cost but helps us produce more educational videos about this amazing universe around us. telescopescanada.ca?bg_ref=KK15sLXmWw
All videos best viewed in 4K.
Peer into the cosmos on Sky Story and along the way come to appreciate our world all the more. Here, we explore the realm of nature beyond our Earth through the fields of astronomy and astrophotography.
All our programs are committed to the standards of scientifically accurate, high quality content that aims to educate and inspire. New videos are posted almost weekly, so please subscribe.
#astronomy #astrophotography #space #nebula #stars
Very interesting info. You have my attention. Thanks for the videos.
It would be interesting to look at the same subject but taken through higher resolution telescopes, and then compare what the drizzled image looks like in comparison to the real higher resolution data. Where does drizzling make the correct assumptions? Where does drizzling go wrong? How can you improve the drizzling process so that it goes wrong less often?
I think some of the answer has to be in how BXT processes the data... I suspect it's doing some kind of adaptive Point Spread Function as it works over the data internally. In fact there's an option to have 'auto PSF' (the default) or not, and choose a manually set fixed PSF. Given the difference between how the non-star data is processed versus the stars data - this seems to indicate what is going on 'under the hood'. Drizzle will change how those adaptive PSFs are derived IMO. A bit like signal processing/filtration on multi harmonic levels in audio ... where interpolating a lower bit rate signal up to a higher rate 'smooths' the data, and thereby creates more 'perfect' signals removing the 'stepping' associated with digital sampling. Similar to wave based sharpening working on different harmonics, to tease out the detail on multiple scales.
It could be. I can't say I have delved much yet into the nuts and bolts of why this works. Too many projects on the go.
thanks for making this video, as drizzle is one of the most greatly misunderstood concepts in the amateur astrophotography community, however this is not a great comparison. its quite obvious you've used some sort of denoise, likely an AI-based one judging by the detailing profile. perhaps noiseX. comparing an algorithm like drizzle on a denoised image is a really inaccurate way to compare them, as you're not able to compare SNR and noise profile/distribution, which is a very fundamental part of the drizzle algorithm. Also as far as I know, there is no benefit in drizzling 2x then downsampling back to a 1x scale, compared to straight drizzling 1x, as the algorithm is just utilising the excessive number of subs to distribute into a finer output grid, which if downsampled defeats the purpose. still, i would ALWAYS recommend drizzling data 1x if not 2x, as its benefits extend way beyond sampling, as covered in-depth in the article from the algorithm developers.
regarding the difference between the two, i would mostly attribute this to differences in processing, as the backgrounds, stretch levels, colours all seem inconsistent throughout your 4 processing attempts. no reason why the post-processing sharpness adjustments or noise adjustments couldn't be variable too. anyways, always lots to learn in this hobby, clear skies!
The development method is covered in the previous video: Upsampling vs Drizzling. The exact same theory of application was applied to develop all the images. BXT-correct. BXT-default. Histogram stretch according to defined parameters defined in my video: Perfect Histograms in 60 Seconds Or Less. NXT at default. Curves adjustment using a two point S curve on the L channel and one point C curve on the S channel applying theory described in my video: Unleash the Power of Curves Transformation. Contrary to myth, it is not possible to develop each image in the exact same way and get fair results as each image carries data altered by its early processing algorithms, thus each will respond differently to the developing steps. It would be like shooting a photo of a subject on Monday at noon using a certain shutter speed, aperture, ISO and deciding the same shutter speed, aperture and ISO will also work for shooting the same subject at noon on Tuesday, Wednesday and Thursday. Application of the same theory of development is a fairer way to approach development for such a test. The differences you see in saturation, brightness and apparent contrast of the outcome images are the result of inherently different qualities in the data after stacking with no changes, stacking and resampling x2, and stacking and drizzling 1x and 2x.
Applying your thoughts on noise, I just re-evaluated the data without running any noise removal. The drizzled images still come out superior in all respects as compared to the correctly sampled but undrizzled data. The apparent signal is stronger and the structures more defined.
The weakness in simply doing a noise assessment to evaluate SNR on these images is SNR has to determine what is background vs noise and this will vary between images. The numbers depend on how well the determination is made. The numbers also don't strictly represent what the human eye perceives. Sometimes one must set aside the numbers and use one's perception.
As Cuiv, the Lazy Geek noted, combining drizzling and BXT is a match made in heaven. Combining both of them with NXT makes the outcomes even better visually. There is no reason other than scientific research that I can think of for not making the most use of the tools at hand.
Thank you for this! At what point to do resample back to the original size? When the image is finished / never?
I don't resample back to normal size. I have tried and found no benefit in doing so other than saving some memory, but memory is cheap these days.
I'd be interested to see the results of this without using BlurXterminator - I wonder if part of the improvement, particularly in the tightness of the stars, is down to the 2X drizzle giving BlurX a higher resolution image to work with.
I think you are fundamentally correct. You might say drizzling is looking for the detail between the pixels. Increasing the size of the canvas probably allows drizzling and BXT to portray the outcomes more clearly. I think I can test this really quickly. I still have the masters.
@@SKYST0RY I'd be keen to see the results of that if you do. I'm not a PixInsight user so I don't have access to BlurXterminator.
I just tested it. Even without BXT, drizzling 2x is distinctly superior on the non-stellar structure and reveals individual stars better in places they interact with non-stellar structure.
@@SKYST0RY Awesome - thanks for doing that.
When applying the 1x stars to the 2x nebulae, do you upscale the stars or downscale the nebulae?
I just upscaled the stars to fit the drizzled x2 plate. There might be a better way. I only recently started experimenting with the technique.
@@SKYST0RY Okay. It's certainly an interesting idea. I'll be giving it a go too. Thanks for sharing.
@@AstroCloudGenerator It works well. I recombine star plates to background plates in Affinity Photo. All I did was drag and stretch the star plate over the bg plate, which upscales it.
Drizzling 1x gives better stars than no drizzle?
You gotta drizzle fo shizzle.
My first glance mind tells me 1 × drizzle looks best. Clear skies.
Okay.
😀