Go Back   IceInSpace > Equipment > Astrophotography and Imaging Equipment and Discussions

Reply
 
Thread Tools Rate Thread
  #1  
Old 11-08-2015, 02:37 PM
Shiraz's Avatar
Shiraz (Ray)
Registered User

Shiraz is offline
 
Join Date: Apr 2010
Location: ardrossan south australia
Posts: 4,787
dither size - fairly long post - corrected version

Hi

this issue came up in Greg's thread - rather than clog that up, thought it might be a good idea to get another going. The following is a summary of where I have got to so far on the topic - be very grateful for any discussion/feedback/results/criticism. thanks for looking. regards Ray

Dither can be used to remove fixed pattern noise (FPN). FPN comes about from a number of sources:
1. pixel-pixel sensitivity variations that modulate sky and target signal.
2. residual noise introduced in calibration, including read noise in flats and darks, shot noise in flats and any dark/bias noise that is impressed on the flats. This residual noise is frozen in the master cal frames and is applied equally to all subs in the calibration process, so it has the same effect as type 1 noise.

The characteristic of FPN is that it is in a pattern that always appears at the same place in the subs – eg if you have an excess signal at a given pixel in one sub due to residual FPN in the master dark, the same excess will appear at that pixel location in all subs and consequently it will still be there after stacking. This is why FPN is so damaging – it does not stack out. However, FPN can be stacked out if you make it uncorrelated between subs and you can do that by dithering the subs, so that the star field moves around between subs. When you align subs on the stars, the FPN is randomly distributed in the subs and FPN behaves the same as any other noise - it reduces when you stack. In fact, if your camera has low pixel-pixel sensitivity variation and you dither, an argument can be made for dispensing altogether with calibration – you can stack out all noise and maybe add some extra dithered subs to compensate for the fact that stacking has to deal with extra noise.

How much do you need to dither? Since the process works best if there is no sub-to-sub correlation in FPN, it seems to make sense that you should ensure that there are more dither positions available than there are subs. For example, if you have 36 subs, it would possibly be wise to have at least a 7x7 dither matrix (49 available positions) in the main image, to ensure that not many subs share star field positions. With random dither, there will almost always be some duplicates, but possibly not to the extent that it will be a major issue unless the FPN is excessive – it may be worth experimenting to see what works best for your system and cal procedures. in particular, there will rarely be a 1:1 correspondence between guide and image sampling, so the above approach will only provide guidance, not an exact solution. And thinking about it, there are probably better ways to dither than random, but that is what current software seems to do.

If you don't have enough dither extent, you will lose some of it's effectiveness and retain a fraction of the FPN. However, any dither will be better than none (even aimpoint drift works) – the last thing you need is a guiding system that places the stars in exactly the same place, sub after sub.

Last edited by Shiraz; 12-08-2015 at 07:58 AM.
Reply With Quote
  #2  
Old 11-08-2015, 04:23 PM
LewisM
Novichok test rabbit

LewisM is offline
 
Join Date: Aug 2012
Location: Somewhere in the cosmos...
Posts: 9,326
Not sure if I am scientifically correct but I dither up to 5 pixels randomly. So far, very little residual noise with the ST8300C and what is left responds to very mild photoshop noise reduction wonderfully (a light Imagenomic pass)
Reply With Quote
  #3  
Old 11-08-2015, 04:33 PM
Atmos's Avatar
Atmos (Colin)
Ultimate Noob

Atmos is offline
 
Join Date: Aug 2011
Location: Melbourne
Posts: 6,339
Great post Ray.
When it comes to dithering and how much, it really depends on what you're doing it for. If you are dithering simply to reduce the FPN then you are better off having the values higher rather than lower (greater offset between subs). This stands to reason as even when doing random dithering there is a lower chance of two subs ended up in the same position.

If you are dithering because you are over sampling and wanting to drizzle more detail out, you then want a lower offset so that the pixel is just moving around a small area. An example of this would be a maximum of 0.8 pixel in any direction from the centre. This leads to each pixel mapping out 1.6 pixels in size with some slight potential overlap of surrounding pixels.

In essence, if all you want to do is reduce noise, it doesn't matter just as long as the subs don't double up. If you intend to drizzle out resolution from over sampling, I personally would make a max of 0.8.

Disclaimer: I haven't been able to test this yet, still waiting for new equipment and clear skies!
Reply With Quote
  #4  
Old 11-08-2015, 04:42 PM
Slawomir's Avatar
Slawomir (Suavi)
Amateur Photon Collector

Slawomir is offline
 
Join Date: Sep 2014
Location: Proserpine
Posts: 3,023
Hi Ray,

I probably should have read the other thread before posting my reply...but here we go...

The degree of dithering needed to minimise residual noise most likely depends on the camera you have, so I would suggest creating quality bias and dark frames and taking time to inspect them.

My understanding is that very few systems would be keeping stars exactly in the same place on the CCD, and even with top end mount you will have sub-pixel variations from sub to sub. Also, I always believed that given the same amount of subs, drizzle integration will result in higher noise in the final image as opposed to non-drizzled integration?
Reply With Quote
  #5  
Old 11-08-2015, 06:25 PM
Shiraz's Avatar
Shiraz (Ray)
Registered User

Shiraz is offline
 
Join Date: Apr 2010
Location: ardrossan south australia
Posts: 4,787
First off, I originally headed the thread as Drizzle, when I meant Dither - seniors moment I guess .

Quote:
Originally Posted by LewisM View Post
Not sure if I am scientifically correct but I dither up to 5 pixels randomly. So far, very little residual noise with the ST8300C and what is left responds to very mild photoshop noise reduction wonderfully (a light Imagenomic pass)
thanks Lewis - whatever works is good . Imagenomic looks to be very capable.

Quote:
Originally Posted by Atmos View Post
Great post Ray.
When it comes to dithering and how much, it really depends on what you're doing it for. If you are dithering simply to reduce the FPN then you are better off having the values higher rather than lower (greater offset between subs). This stands to reason as even when doing random dithering there is a lower chance of two subs ended up in the same position.

If you are dithering because you are over sampling and wanting to drizzle more detail out, you then want a lower offset so that the pixel is just moving around a small area. An example of this would be a maximum of 0.8 pixel in any direction from the centre. This leads to each pixel mapping out 1.6 pixels in size with some slight potential overlap of surrounding pixels.

In essence, if all you want to do is reduce noise, it doesn't matter just as long as the subs don't double up. If you intend to drizzle out resolution from over sampling, I personally would make a max of 0.8.

Disclaimer: I haven't been able to test this yet, still waiting for new equipment and clear skies!
Very good points Colin - thanks. Despite an original heading indicating that the post was about drizzle, my post was based on dither in terms of imager pixels and the assumption is that dither will be in whole pixels. I will modify the first post. What I was trying to do was provide a justification for broader dither than people may be using by showing why it could possibly help. A user will need to translate the guider dither into imager dither to find any value in the suggestions.

Dither for super-resolution requires sub-pixel misalignment. After registration, 2.3 pixels or 7.3 pixels offset will be functionally equivalent to 0.3 pixels, so I suspect that dither can be arranged to be appropriate for both noise reduction and super-resolution. As Slawomir points out, it is reasonable to assume that any guidance system will produce some degree of mismatch anyway. In a previous life, did some effective super-res experiments with nothing more than linear steps at a scale that was much larger than the main imaging pixels, but not harmonic - it worked fine on much larger than sub-pixel steps.

Quote:
Originally Posted by Slawomir View Post
Hi Ray,

I probably should have read the other thread before posting my reply...but here we go...

The degree of dithering needed to minimise residual noise most likely depends on the camera you have, so I would suggest creating quality bias and dark frames and taking time to inspect them.

My understanding is that very few systems would be keeping stars exactly in the same place on the CCD, and even with top end mount you will have sub-pixel variations from sub to sub. Also, I always believed that given the same amount of subs, drizzle integration will result in higher noise in the final image as opposed to non-drizzled integration?
Thanks Slawomir. Dither is an extra tool that can either replace or enhance standard calibration. Having high quality calibration data is essential if you rely on conventional calibration, but it can possibly become a burden as much more cal data is needed with deep imaging - dither can reduce the need for massive cal datasets and it deals effectively with any residual noise (eg shot noise from dark current outliers), that will persist even in large cal datasets.

Agree that any guide system will not keep the stars exactly in place, but some come close. The point that I was trying to make is that, even though it seems like a laudable goal, trying to get the guiding so good that stars stay in the same place on the CCD in all of the subs is actually counter-productive.

Dither can definitely help with noise reduction when used with conventional stacking - I stuffed up the heading to the thread and used the word drizzle when I meant dither, sorry about that..

Last edited by Shiraz; 12-08-2015 at 07:01 AM.
Reply With Quote
  #6  
Old 12-08-2015, 06:59 AM
Slawomir's Avatar
Slawomir (Suavi)
Amateur Photon Collector

Slawomir is offline
 
Join Date: Sep 2014
Location: Proserpine
Posts: 3,023
No worries Ray. My comment was in regards to drizzle and the need for have some movement between subs for it to work properly, so it can be totally ignored
Reply With Quote
  #7  
Old 12-08-2015, 12:21 PM
PRejto's Avatar
PRejto (Peter)
Registered User

PRejto is offline
 
Join Date: Jan 2011
Location: Sydney, Australia
Posts: 1,291
Hi Ray,

I brought this up in Greg's post so I appreciate the insights posted by all in this thread. I have to admit that I had not fully considered the difference between the guide camera move (in pix) vs the imaging camera. Duh. Fortunately given that my guide camera's pix are larger than the imaging camera I think it would be safe to assume that a 1 pix move of the guide camera is a greater move on the imaging camera. But, that raises a different thought to puzzle out; if the guide camera is binned 2x2 and one moves by 1 pix is that the same as moving 4 pix (or 2?) if the camera was not binned?

Peter
Reply With Quote
  #8  
Old 12-08-2015, 01:09 PM
gregbradley's Avatar
gregbradley
Registered User

gregbradley is offline
 
Join Date: Feb 2006
Location: Sydney
Posts: 15,473
Doesn't this also assume perfect tracking with no drift at all between subs. Is that the case with your system? Dithering implies every sub is exactly the same without a dither command. I rarely see that even with high end mounts.

If you work your PA to perfection maybe but if you are using a higher PE mount I think it highly unlikely to the extent of 1 or 2 pixels or even more.

Another point not quite related is its common practice to bin your guider 2x2 as Peter pointed out. But if you can I think there is a gain by using 1x1 binning. Perhaps not practical if its hard to find a guide star.

SBIG STi has 7.4 micron pixels. Not sure about a Lodestar. 2x2 binned would 7.4 micron pixels would be the equivalent of 15 micron pixels. So if you are using a 9 micron pixelled camera you are already over 1.6 pixels. I suppose the guide star would land within that 15 pixel band at optimum but highly unlikely it would in practice and no doubt lands somewhere within several pixels. Even with good seeing sitting watching the guide star it changes shape all the time so the centroid would have to be moving around as well.

My point is even if you aren't deliberately dithering there would still be a significant amount of dithering occurring anyway as mounts PE, polar alignment drift, flexes and seeing are causing it to move around already.


Otherwise you wouldn't need a Protrack TPoint model if it were dead still sub after sub and within a sub.

But as an approach this is good work and it makes sense to make subs not replicate each other exactly.


Greg.
Reply With Quote
  #9  
Old 12-08-2015, 03:29 PM
Paul Haese's Avatar
Paul Haese
Registered User

Paul Haese is offline
 
Join Date: Jan 2009
Location: Adelaide
Posts: 9,430
Ray I am using +/- 1.3 pixels of dither via guider with a maximum guide error of +/- 1 pixel with a cycle of 10 in CCDAP. What I effectively get is a little dance around with all my subs. Added to this is that I do 3 precision slews after each focus run to 23 arc seconds only, it mainly gets to 7 arc seconds. That effectively means I don't get any real overlap from one sub to another but sometimes get the occasional sub that does correlates with an earlier one. I am not sure if I am dithering enough though the total effect of pointing error with dither should be enough. Thoughts?
Reply With Quote
  #10  
Old 12-08-2015, 04:21 PM
gregbradley's Avatar
gregbradley
Registered User

gregbradley is offline
 
Join Date: Feb 2006
Location: Sydney
Posts: 15,473
I think Peter's problem is the Sky X does dither from the camera control tab not the autoguider. I imagine that would be the same as getting a sudden spike in PE and how long does it take for the guiding to settle back down. Maxim must do it differently. It sounds like SB chose the weaker method.

Greg.
Reply With Quote
  #11  
Old 12-08-2015, 05:22 PM
Shiraz's Avatar
Shiraz (Ray)
Registered User

Shiraz is offline
 
Join Date: Apr 2010
Location: ardrossan south australia
Posts: 4,787
Quote:
Originally Posted by PRejto View Post
Hi Ray,

I brought this up in Greg's post so I appreciate the insights posted by all in this thread. I have to admit that I had not fully considered the difference between the guide camera move (in pix) vs the imaging camera. Duh. Fortunately given that my guide camera's pix are larger than the imaging camera I think it would be safe to assume that a 1 pix move of the guide camera is a greater move on the imaging camera. But, that raises a different thought to puzzle out; if the guide camera is binned 2x2 and one moves by 1 pix is that the same as moving 4 pix (or 2?) if the camera was not binned?

Peter
Hi peter. I think that 2x2 binning gives twice the movement in the image. If the guide pix are bigger than the imaging pix, I guess you get more motion in the image.

Quote:
Originally Posted by gregbradley View Post
Doesn't this also assume perfect tracking with no drift at all between subs. Is that the case with your system? Dithering implies every sub is exactly the same without a dither command. I rarely see that even with high end mounts.

If you work your PA to perfection maybe but if you are using a higher PE mount I think it highly unlikely to the extent of 1 or 2 pixels or even more.

Another point not quite related is its common practice to bin your guider 2x2 as Peter pointed out. But if you can I think there is a gain by using 1x1 binning. Perhaps not practical if its hard to find a guide star.

SBIG STi has 7.4 micron pixels. Not sure about a Lodestar. 2x2 binned would 7.4 micron pixels would be the equivalent of 15 micron pixels. So if you are using a 9 micron pixelled camera you are already over 1.6 pixels. I suppose the guide star would land within that 15 pixel band at optimum but highly unlikely it would in practice and no doubt lands somewhere within several pixels. Even with good seeing sitting watching the guide star it changes shape all the time so the centroid would have to be moving around as well.

My point is even if you aren't deliberately dithering there would still be a significant amount of dithering occurring anyway as mounts PE, polar alignment drift, flexes and seeing are causing it to move around already.


Otherwise you wouldn't need a Protrack TPoint model if it were dead still sub after sub and within a sub.

But as an approach this is good work and it makes sense to make subs not replicate each other exactly.


Greg.
Hi Greg. The point I was trying to get at is that dither may well have to be much larger than a few pixels to be fully effective. Mount errors etc will introduce some FPN decorrelation sub-to-sub and you could dial up a couple of pixels dither as well, but it could be that 3-4 pixels is nowhere near enough.

Quote:
Originally Posted by Paul Haese View Post
Ray I am using +/- 1.3 pixels of dither via guider with a maximum guide error of +/- 1 pixel with a cycle of 10 in CCDAP. What I effectively get is a little dance around with all my subs. Added to this is that I do 3 precision slews after each focus run to 23 arc seconds only, it mainly gets to 7 arc seconds. That effectively means I don't get any real overlap from one sub to another but sometimes get the occasional sub that does correlates with an earlier one. I am not sure if I am dithering enough though the total effect of pointing error with dither should be enough. Thoughts?
Sounds like it should be good Paul, provided you do regular refocus/slew. Even occasional correlation will add measurable noise though.

Quote:
Originally Posted by gregbradley View Post
I think Peter's problem is the Sky X does dither from the camera control tab not the autoguider. I imagine that would be the same as getting a sudden spike in PE and how long does it take for the guiding to settle back down. Maxim must do it differently. It sounds like SB chose the weaker method.

Greg.
This is another good issue - it is all very well to dither by a lot, but it may be a problem if the mount takes minutes to settle down again.

Thanks for the discussion. when Peter posted his question, I realised that I did not know how much dither was enough. A rough and ready analysis suggested that I should be using larger dither and it is good to get other opinions and critique - the analysis was clearly a bit too simplistic. Has anyone done any tests by any chance - I will try to do so next time the sky clears (don't hold your breath).

regards Ray

Last edited by Shiraz; 12-08-2015 at 07:09 PM.
Reply With Quote
  #12  
Old 12-08-2015, 05:26 PM
PRejto's Avatar
PRejto (Peter)
Registered User

PRejto is offline
 
Join Date: Jan 2011
Location: Sydney, Australia
Posts: 1,291
Quote:
Originally Posted by gregbradley View Post
I think Peter's problem is the Sky X does dither from the camera control tab not the autoguider. I imagine that would be the same as getting a sudden spike in PE and how long does it take for the guiding to settle back down. Maxim must do it differently. It sounds like SB chose the weaker method.

Greg.
I don't thnk it matters where the dither command originates. It just tells the mount to move x arc-sec between exposures. The guide camera never stops so it just works out the correction and the next exposure begins when the guider error gets below a certain user adjustable threshold. The operative word is "between" exposures so it wouldn't be a spike or anything ever seen in the image.

Peter
Reply With Quote
  #13  
Old 12-08-2015, 05:55 PM
Peter.M's Avatar
Peter.M
Registered User

Peter.M is offline
 
Join Date: Sep 2011
Location: Adelaide
Posts: 914
With SGP plate solving my images and putting the mount in the "same" place for a second night I found that I would have very little fixed pattern noise after a few nights. And the biggest eliminator of FPN that I have found is meridian flips with no rotator.
Reply With Quote
  #14  
Old 12-08-2015, 06:21 PM
Shiraz's Avatar
Shiraz (Ray)
Registered User

Shiraz is offline
 
Join Date: Apr 2010
Location: ardrossan south australia
Posts: 4,787
Quote:
Originally Posted by Peter.M View Post
With SGP plate solving my images and putting the mount in the "same" place for a second night I found that I would have very little fixed pattern noise after a few nights. And the biggest eliminator of FPN that I have found is meridian flips with no rotator.
I guess that is consistent with the idea that large dither is best. Meridian flip is essentially extra large dither - it completely removes FPN correlation, but you could get the same effect by moving the subs in a large dither pattern. Do you use dither?

Last edited by Shiraz; 12-08-2015 at 07:10 PM.
Reply With Quote
Reply

Bookmarks

Thread Tools
Rate This Thread
Rate This Thread:

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT +10. The time is now 03:42 AM.

Powered by vBulletin Version 3.8.7 | Copyright ©2000 - 2019, Jelsoft Enterprises Ltd.
Advertisement
OzScopes Authorised Dealer
Advertisement
Celestron Australia
Advertisement
Meade Australia
Advertisement
SkyWatcher Australia
Advertisement
Bintel
Advertisement
NexDome Observatories
Advertisement
Lunatico Astronomical
Advertisement
Astronomy and Electronics Centre
Advertisement