Under sampling problem?

[DougM43]

Hi all,

its taken me a while but I have just discovered the ccd compatibility tool on the forum and have checked my camera with my imaging scope (skywatcher ed80 pro and starlight express sxvr m25c and it shows it is under sampling by quite some margin (even worse if I include the .85 reducer)

ive heard people say under or over sampling is no big deal but this really doesn’t look good to me.

 

ive spent quite a bit of time this weekend inputting various cameras into the calculator and the ZWO asi 1600mm pro seems a perfect match for the ed80, but before I press the buy it button on a £1500 camera and filters I thought it wise to check if there’s any other solution to the problem or if  indeed there is  a problem at all.

thanks in advance, Doug.

[cletrac1922]

Doug

The ZWO asi 1600mm is top end of the ZWO range, and also a cooled camera as well

The atached link is presented by one of my club members

 

 

John

 

 

[michael8554]

Previous posts on sample rate will take you several days to read, it's a very contentious subject..

Are you happy with your results.? 

That's all that really matters. 

Michael 

[andrew s]

Google "drizzle imaging" it is a technique for under sampled images.

Regards Andrew 

[vlaiv]

18 minutes ago, andrew s said:

Google "drizzle imaging" it is a technique for under sampled images.

Regards Andrew 

Why everyone keeps saying do the drizzle, do the drizzle - when it does not work with amateur setups

In order for drizzle to work, one must have very accurate pointing (guiding) - and needs to control dither to exact sub pixel precision. This works for Hubble (origin of technique) because it can be pointed with adequate precision and does not need guiding.

For drizzle to work, it needs to be applied to basic shift & add stacking. This means that there should be no interpolation of samples to align frames in registration / alignment phase of stacking. Drizzle is a technique that goes "beyond" simple shift and add and allows for fraction of pixel shift - but strictly defined, like half a pixel or 1/3.

When one tries to do drizzle in amateur setup - you have random shifts between subs, you need to use interpolation (and depending on type of interpolation all sorts of things happen with the data - there is inevitable correlation between pixels introduced that leads to even more blur instead of bringing out the details, or if more sophisticated interpolation is used - there will be band pass limit on sampling frequency to avoid aliasing).

You end up "thinning" your data with "drizzle" - without true benefit of drizzle - one might as well use up sampled stacking - it will probably produce better results.

Dealing with oversampling is really easy - use barlow to get to focal length / sampling rate that you find suitable. Now people will start yelling - you will "slow down" your scope that way, and yes you will, but ED80 at 1.5"/px will not be slower than another ED80 at 1.5"/px (one using camera with smaller pixels).

[ollypenrice]

So you're at 3.1"PP?

Tom O'Donoghue and I were runners up in the Astrophotographer of theYear competition with an image shot at 3.5"PP. There are dozens of APODS shot with the Kodak 11 meg chip and the Tak FSQ106 at 3.5"PP. Here are some images taken at this scale:

https://www.astrobin.com/327970/?nc=user

https://www.astrobin.com/301531/?nc=user

https://www.astrobin.com/383965/?nc=user

Don't panic, Mr Mannering! 

Olly

 

[andrew s]

8 hours ago, vlaiv said:

Why everyone keeps saying do the drizzle, do the drizzle - when it does not work with amateur setups

Others seem to differ http://www.astrosurf.com/buil/us/spe9/lrgb22.htm example with real amateur data.

Regards Andrew

[vlaiv]

2 hours ago, andrew s said:

Others seem to differ http://www.astrosurf.com/buil/us/spe9/lrgb22.htm example with real amateur data.

Regards Andrew

I'm aware that people think it's working, and indeed it is working under certain circumstances. My point is that under amateur setups, basic preconditions for it are not honored, and thus it will not be working as expected.

The page you quoted, explains what happens when you have irregular sub pixel shifts. You end up splitting pixel by surface, and assigning a part of pixel to each cell. This is in effect very similar to doing bilinear interpolation which introduces more blur and correlation between sample values.

For in depth explanation of what I mean:

http://www.stsci.edu/itt/DocsStudyGroup/MultiDrizzle/html/ch3.5.2.html

Pay particular attention to following:

Interlace method is the one I was describing in my post above (very accurate sub pixel shifts), authors here say that it is not feasible even for Hubble - it needs exact sub pixel dithers between images.

Shift and add method:

Quote

Shift-and-add has the advantage of being able to easily handle arbitrary dither positions. However, it convolves the image yet again with the original pixel, thus adding to the blurring of the image

Drizzle method:

Quote

The degree to which the algorithm departs from interlacing and moves towards shift-and-add depends upon how well the PSF is sub-sampled by the shifts in the input images.

Or rather, in amateur terms, we don't talk about optics PSF, but combined system PSF (which happens to be different between subs due to seeing effects and guiding errors). For drizzle to depart from shift and add and approach interlacing (thus having any chance of improvement rather than just additionally blurring things out) - you need to be able to control subsampling of PSF, or you need to be able to control your sub pixel dither with any sort of precision.

If it is random, and you have different PSF in each sub - you will just end up being on shift and add side of "scale" and result will just be more blurred (like I explained above - similar to what bilinear interpolation does).

[andrew s]

However  @vlaiv, the processed images are an improvement. While you may be correct from a certain theoretical perspective the process, what ever you might want to call it works in the example gi en.

Regards Andrew 

[vlaiv]

14 minutes ago, andrew s said:

However  @vlaiv, the processed images are an improvement. While you may be correct from a certain theoretical perspective the process, what ever you might want to call it works in the example gi en.

Regards Andrew 

You mean this example (from Buil's page you linked to)? :

Here is counter example that I did in like 1 minute:

Just because left image looks pixelated and the right one does not - that does not mean that there is actual recovery of detail using this method. Image is misleading because left image was enlarged to match scale of right image by using nearest neighbor filtering. I produced my example by using same image - one was just enlarged (nearest neighbor) to show "pixels" while other one was rescaled using other interpolation method.

Although right image looks "smoother" and with "more resolution" - it is in fact the same image.

True test for drizzle in amateur setups can be easily performed. One should take any set of dithered images, and bin them (in software) to get to stage where there is adequate under sampling. One should then proceed as follows:

- do drizzle stacking of such set

- do comparison stack that will be made by scaling up each sub by drizzle factor prior to stacking, and then stack those subs.

Compare two for resolution - by both visual inspection (and while at it - check SNR of each) and by means of star FWHM in each image. One with lower FWHM should be "more resolved".

[Pompey Monkey]

59 minutes ago, vlaiv said:

You mean this example (from Buil's page you linked to)? :

Here is counter example that I did in like 1 minute:

Just because left image looks pixelated and the right one does not - that does not mean that there is actual recovery of detail using this method. Image is misleading because left image was enlarged to match scale of right image by using nearest neighbor filtering. I produced my example by using same image - one was just enlarged (nearest neighbor) to show "pixels" while other one was rescaled using other interpolation method.

Although right image looks "smoother" and with "more resolution" - it is in fact the same image.

True test for drizzle in amateur setups can be easily performed. One should take any set of dithered images, and bin them (in software) to get to stage where there is adequate under sampling. One should then proceed as follows:

- do drizzle stacking of such set

- do comparison stack that will be made by scaling up each sub by drizzle factor prior to stacking, and then stack those subs.

Compare two for resolution - by both visual inspection (and while at it - check SNR of each) and by means of star FWHM in each image. One with lower FWHM should be "more resolved".

Increased information or not, I know which image looks better to me.

[vlaiv]

14 minutes ago, Pompey Monkey said:

Increased information or not, I know which image looks better to me.

Quite so! That is why we use better interpolation techniques for resampling images.

Problem with drizzle however is that you reduce SNR of your images to recover detail (sparse samples, and each resulting pixels gets less samples stacked so SNR gain is lower).

Let's suppose that there are tangible gains in resolution by using drizzle method in amateur setups. These gains won't be 100% as there is always pixel blur introduced. Drizzle lowers total SNR by the same factor as spread image over more pixels (that is what it is doing after all). Question now is, why not use barlow? Effect is the same - lower SNR because light is spread over more pixels, but there is no longer need to correct for undersampling and pixel blur (which even if it works, won't be 100%).

Drizzle was invented for Hubble, for two reasons - first it has predictable PSF and pointing accuracy is high (one can do very close to exact fractional shifts). Second, of course was the fact that swapping camera for one with smaller pixels, or adding a barlow lens (focal extender) was not something you could easily do.

If gains in resolution are minimal (shift and add side of things) - one would be well advised to use up scaled stacking (just resample each sub by certain factor - for example x2 and then stack those) as this will not lead to SNR loss. If one on the other hand wants to go for proper sampling (or less undersampling) - there are two options, more expensive one - get camera with smaller pixels, and really cheap one - get a barlow.

Planetary imagers are well aware that they lack cameras with sufficiently small pixels in most cases to get to critical sampling rate, and have no problem using barlows. Why don't they just drizzle instead? (just a remark, bayer drizzle is something different and it actually works because pixels are already "smaller" / sparse and there is no need to "shrink" them - which requires particular offset when trying to properly sample PSF). Simple answer to that one - barlow just works.

[andrew s]

No image processing  can recover more information than is in the original captures. All it can do is highlight some aspects more than others or present it in a more pleasing form.

Image processing just trades off various factor to suit our visual systems and the eye of the processor.

Yes drizzle is not perfect even with Hubble.

Yes you can use a Barlow but it reduces the field and adds its own aberrations.

Regards Andrew

[vlaiv]

Just to wrap up this side tracking of the thread (sorry OP, but as was pointed out - discussion about under / over sampling gets crazy )

This is montage of two crops (same area) of drizzle (upper) and resampled stacking (lower) of the same data.

I took 64 subs of Ha subs on Bubble nebula. Subs are calibrated and binned to 1"/px. FWHM measurements show that most stars are around 3.3".

Subs were binned x3 (average in software) - all that was done to obtain first set for stacking. I opted for x3 binning as that would put data at "proper" under sampling for drizzle - as per Buil's page FWHM around 1px.

This set was stacked using DSS and drizzle x3 (regular average, no background calibration, no cosmetic correction, just plain regular stacking with drizzle x3).

Original set was then resampled x3 using Quintic B-Spline (TransformJ plugin of ImageJ). This created second set of subs. These were stacked in DSS with the exact same parameters as first (same ref frame and options), except for drizzle - which was turned off.

Mosaic is only linearly stretched and no further processing was done.

To my eye, lower image displays greater resolution, and only suffers from ringing artifacts associated with resample method used - Quintic B-Spline (don't have Lanczos filter implementation yet for rescaling).

Upper drizzle image has poorer resolution and also, DSS struggled to align stars properly and there is distinct elongation of stars in bottom left - upper right direction (probably limitation of DSS implementation of algorithm?).

I guess that above example is pretty indicative of how "successful" drizzle is in amateur setups.

 

[andrew s]

3 minutes ago, vlaiv said:

I guess that above example is pretty indicative of how "successful" drizzle is in amateur setups.

 

No. It show how drizzel works with you simulation. Binning is not the same as dithering at lower resolution.  However, I am not sure thus discussion is very productive. 

I know you have strong views back by considerable theoretical understanding I accept we reach different conclusions. 

Regards Andrew 

[vlaiv]

5 minutes ago, andrew s said:

No. It show how drizzel works with you simulation. Binning is not the same as dithering at lower resolution.  However, I am not sure thus discussion is very productive. 

I know you have strong views back by considerable theoretical understanding I accept we reach different conclusions. 

Regards Andrew 

Sure, we need not pursue this any further, however, for clarity, I would like to state that Binning - as in terms of straight average of adjacent pixels of already dithered set of images (multi pixel offsets) is indeed equivalent of using camera with larger pixels (light adds up) with sub pixel dithering. All offsets in DSS were fractional and each sub had some offset compared to reference frame when registration of binned set was performed.

[DougM43]

Wow, I feel quite humbled at some of the replies, the level of knowledge and experience given freely on this forum is quite astounding, I just hope one day I’m in a position to help other beginners move their hobby forwards.

@cletrac1922 thank you for the link, I actually watched that same video along with many others regarding ZWO cameras on Saturday, after reading some of the other replies I’m going to stick with my Current camera for the next one or maybe two winters before going mono, I must admit I particularly like the idea of the ASIair, setting up an imaging rig and running it via my iPad instead of all the trailing wires to a laptop is quite appealing.

@andrew s and @vlaiv I don’t pretend to fully understand your interesting discussion but it has taught me I have so much more to learn, I spent pretty much of what limited imaging time last winter getting to grips with APT and guiding with PHD so hopefully this winter can be spent actually  collecting data, 

vlaiv you mentioned using a Barlow, one of my targets to get a reasonable image of is M51, but on my camera it is incredibly small and when I crop it to a reasonable size it definitely looses resolution, if I understand correctly using a Barlow on small targets such as this would be a good solution to a bigger better final image?

@ollypenrice thank you that is just what I wanted to hear, I have seen plenty of good images with the ed80 and I have seen plenty of good images from the sxvr camera but I was thinking I’d got a roller skate on one foot and an ice skate on the other,, both great at what they do but hopeless together, you’ve put my mind at ease, those images you posted are superb 

panic over Mr Mainwaring. 🙂

[vlaiv]

9 minutes ago, DougM43 said:

vlaiv you mentioned using a Barlow, one of my targets to get a reasonable image of is M51, but on my camera it is incredibly small and when I crop it to a reasonable size it definitely looses resolution, if I understand correctly using a Barlow on small targets such as this would be a good solution to a bigger better final image?

Actually, the solution that I would personally go with would be getting another scope for such targets. Very nice scope that would be well matched to that camera for M51 and smaller targets would be 6" RC.

However, given that you are still climbing the learning curve, such scope might not be the best solution. It is reflector type scope and it requires a bit of fiddling with to get it working properly. Collimation can be tricky and in all likelihood it would benefit from focuser upgrade.

Probably easiest way to get higher resolution images would be to use a barlow. In this case you should aim for 1.5 - 2.0"/px sampling resolution. This means focal lengths between 800mm and 1050mm, or barlow multiplier of x1.33 to x1.75.

Good thing about barlows is that you can change magnification factor of it by changing distance to sensor (not so with focal extenders / telecentrics / powermates). Given the size of your camera sensor, you will need a bigger barlow. One that is certainly top tier and very well suited for this purpose is Baader VIP barlow if the price is suitable (it is expensive compared to most other units). It is designed to illuminate larger sensors and it is modular, so you can attach it in different ways. I believe barlow element has t2 thread on it so you can use only that and provide your own spacing. There is a barlow magnification formula - that goes 1+ (distance/barlow focal length). Only thing that you need to know is barlow focal length (65.5mm in case of VIP barlow) and you can calculate need distance to get required magnification factor.

Size of M51 when viewed at full resolution (1:1 - meaning one image pixel to one screen pixel) with 2"/px sampling will be:

With sampling of 1.5"/px this would be the size:

You might be tempted to jump straight to 1.5"/px but in order to utilize such resolution you will need steady skies and good guiding. 80mm aperture is rather small and it has less resolving power than larger scopes. This combined with seeing and guiding precision will limit useful resolution to about 2"/px maybe a bit less on most nights.

One more thing - more you magnify your image, more total imaging time you will need to get the same SNR, so it's probably best to start out at 2"/px and when you feel confident - push it a bit more towards 1.5"/px

HTH

 

[DougM43]

Vlaiv, thanks again for your in-depth and very helpful answer, you have given me a lot to think about,

 

[michael8554]

Your LX90 with a Focal Reducer would be a good match? 

Michael 

[vlaiv]

58 minutes ago, michael8554 said:

Your LX90 with a Focal Reducer would be a good match? 

Michael 

Actually it would, even without focal reducer.

I would not try to guide such a scope without OAG though. And processing would be somewhat more involved - at least if running without focal reducer. I would personally exploit the fact that Bayer matrix on OSC is actually every other pixel for each color - that would mean 7.8um at 1000mm (although scope is 2000 mm - if it is 8" version). That would put it at very comfortable 1.61"/px with 8" aperture.

[DougM43]

Thanks again I don’t really want to de-fork my lx90 to mount it on an eq mount, 

i travel up to the Orkney isles two or three times a year to visit family and take the lx with me and when the weather permits spend a bit of time with my father star gazing under some truly dark skies, the speed and ease of setting it up makes it not quite but almost grab and go.

i think I will order a baader vip Barlow and go with that suggestion, I can see that working well.

thanks again,, very helpful 

Doug.

[DougM43]

The baader Vip Barlow has now arrived and I thought it would be quite self explanatory as how to fit it into the imaging train but alas not,, the problem as I see it is the Barlow element ( the glass bit) is in the front 1.25” bit so I need some way to attach it to the 2” telescope  draw tube,, short of sliding it into the telescopes diagonal, camera an all, I can’t see what else I can do with it,

the picture shows the camera and corrector as I have it on the scope and the Barlow so you can hopefully see what the problem is.

anybody got ant ideas??

I’m assuming with the Barlow I won’t be using the reducer/ corrector

Doug

Edited June 27, 2019 by DougM43

[DougM43]

Right, after a nice cup of tea I think I’ve got this sorted, 

i have unscrewed the baader Vairilock extension and reducer from the draw tube, they have come off together which is good as it will preserve the distance from the reducer to the camera sensor when I come to use it again.

the new Barlow has screwed directly onto the front of the camera, just had to remove the two knurled locking screws meant for locking in an eyepiece to allow it to screw right up.

replaced the black adapter meant for holding the diagonal onto the draw tube and the baader Vip  slides right in.

i haven’t measured it but it looks like the Barlow element is sitting pretty much in the centre of the focus tube travel so hopefully it will come to focus ok with the camera.

with the Barlow itself I have removed one of the 15mm T2 extension pieces so it now has a total length of 67mm to the sensor so effectively doubling the telescope focal length.

a couple of pictures explain it better 

many thanks, Doug.

 

 

Edited June 27, 2019 by DougM43

[vlaiv]

I was about to recommend that you try something along those lines - glad you sorted it easily.