25 replies to this topic

[thomas7]

Has anyone had a similar problem? I added drizzle in WBPP with settings 2x drizzle, drop shrink 0.9 and VarShape 1.5.

The final image looks good, but when I use drizzle integration with the same settings, these strange artifacts appear.

I tried background extraction, blur exterminator, SPCC but nothing helped.

If anyone has any ideas or suggestions, please help.

 

The camera is a ZWO ASI294MC Pro, and the telescope is a William Optics ZS61. The resolution is 2.65 arcs/pixel

 

[idclimber]

How much data and how much dithering? If you do not have enough different dithered locations, drizzle integration will not work well at all.

[thomas7]

Night 1: 50 exposures 
Night 2: 40 exposures 
Night 3: 20 exposures 
Night 4: 35 exposures 
Total: 7 hours 
Dither: Every 5 frames, 5-pixel offset

All frames listed are approved only from using Subframe Selector

[idclimber]

That is only 145 frames. This divided by 5 is 29 unique locations. You need a minimum of double that. I would dither every other frame as a minimum for that sized data set. 

 

You also need to verify you are under sampled. What is the fwhm of a raw sub exposures. Also the eccentricity. Since you apparently have PixInsight you can measure this using the FWMHEcentricity script. 

[idclimber]

You also need to be aware that drizzle 2x will decrease your SNR by about a factor of two. This is why this requires a ton of data. In fact about four times the total integration time to make up for the loss of SNR. 

[thomas7]

I used Dylan's Astronomy Setup Calculator to see if I'm undersampled. 

I did not know that this amount of data is needed for. I did my research too poorly obviously. 

 

Here are my results for fwhm and eccentricity

[psienide]

Has anyone had a similar problem? I added drizzle in WBPP with settings 2x drizzle, drop shrink 0.9 and VarShape 1.5.

The final image looks good, but when I use drizzle integration with the same settings, these strange artifacts appear.

I tried background extraction, blur exterminator, SPCC but nothing helped.

If anyone has any ideas or suggestions, please help.

 

The camera is a ZWO ASI294MC Pro, and the telescope is a William Optics ZS61. The resolution is 2.65 arcs/pixel

 

orion drizzle pattern.jpg

orion drizzle pattern zoom.jpg

Did you make sure to use CFA drizzle? If so - it has to be done on undebayered images as far as i'm aware. I'm not sure how PI handles this with WBPP, but I saw similar artifacts when doing this the first time in siril.

[idclimber]

The calculator will tell you little.

 

Your fwhm in pixels is about what I imaged with last night with an image scale of 0.86"/px. This is a focal length of 900mm on my 6" SVX152.  This is about at the limits of my seeing.

 

Here is a screenshot of one of the subs. Notice how round the stars are. This is not under sampled and would not benefit from drizzle. At least not without a ton of data and not my much. If you inspect one of your subs, my bet is they are very comparable. Signs of under sampled would be square stars. 

 

 

Your fwhm is larger than I would expect at that image scale. I would look at your focus and perhaps guiding. I am less inclined to consider guiding given your good eccentricity. 

[psienide]

That is only 145 frames. This divided by 5 is 29 unique locations. You need a minimum of double that. I would dither every other frame as a minimum for that sized data set. 

 

You also need to verify you are under sampled. What is the fwhm of a raw sub exposures. Also the eccentricity. Since you apparently have PixInsight you can measure this using the FWMHEcentricity script. 

What's your basis for this? If it's just noise, why does the number of dithers matter? Just out of curiosity

I drizzle almost everything and just want make sure i'm not missing something. I do every 3 on my 5min NB captures (averaging 100 integrated frames after QC), and every 5 on my BB.

[NMCN]

That is only 145 frames. This divided by 5 is 29 unique locations. You need a minimum of double that. I would dither every other frame as a minimum for that sized data set. 

 

You also need to verify you are under sampled. What is the fwhm of a raw sub exposures. Also the eccentricity. Since you apparently have PixInsight you can measure this using the FWMHEcentricity script. 

I've never heard of this before, but would like to hear more about it. Is this the new thinking on dithering? I've always gone by how long my subs are. 90 seconds or less and I dither every 3 frames. 91 to 120 seconds, I dither every 2 frames. Longer than 180 seconds, I dither every frame. Never had an issue. You also need to make sure your dithering is using a random pattern vice a spiral pattern.

 

All my thinking is based on this older post. Maybe the info is out of date now. I don't know.

[thomas7]

Did you make sure to use CFA drizzle? If so - it has to be done on undebayered images as far as i'm aware. I'm not sure how PI handles this with WBPP, but I saw similar artifacts when doing this the first time in siril.

Yes I did. As far as I know, CFA drizzle is performed before debayering.

[psienide]

I've never heard of this before, but would like to hear more about it. Is this the new thinking on dithering? I've always gone by how long my subs are. 90 seconds or less and I dither every 3 frames. 91 to 120 seconds, I dither every 2 frames. Longer than 180 seconds, I dither every frame. Never had an issue. You also need to make sure your dithering is using a random pattern vice a spiral pattern.

 

All my thinking is based on this older post. Maybe the info is out of date now. I don't know.

Seems like we need to work out what's the most efficient strategy for dithering with intent to drizzle. I don't seem to have any problems with dithering every 3 frames on 5 min captures. But if dithering every frame will afford some additional benefit I would switch.  

[psienide]

Yes I did. As far as I know, CFA drizzle is performed before debayering.

And you started with all the raw frames? Again, not a PI guy so not sure if it's possible - but it's easy to make mistakes in Siril. If I go back and try to drizzle already registered frames it gives me weirdness.

[idclimber]

What's your basis for this? If it's just noise, why does the number of dithers matter? Just out of curiosity

I drizzle almost everything and just want make sure i'm not missing something. I do every 3 on my 5min NB captures (averaging 100 integrated frames after QC), and every 5 on my BB.

I can't point you to a specific thread on this without spending time searching this forum. My statement references dozens of discussions on this topic over the last several years paying attention to those advanced imagers that I trust. 

 

The concept is simple, you need intermediate data for this to work. That is you need data from 1/3 and 1/2 pixel over from the reference frame. This is done in a circular pattern around the reference frame. 

 

 

I've never heard of this before, but would like to hear more about it. Is this the new thinking on dithering? I've always gone by how long my subs are. 90 seconds or less and I dither every 3 frames. 91 to 120 seconds, I dither every 2 frames. Longer than 180 seconds, I dither every frame. Never had an issue. You also need to make sure your dithering is using a random pattern vice a spiral pattern.

 

All my thinking is based on this older post. Maybe the info is out of date now. I don't know.

There are two reasons to dither. The primary one is move the fixed pattern noise around so it is eliminated by stacking. The second is to drizzle integrate 2x if your data supports it. 

[thomas7]

The calculator will tell you little.

 

Your fwhm in pixels is about what I imaged with last night with an image scale of 0.86"/px. This is a focal length of 900mm on my 6" SVX152.  This is about at the limits of my seeing.

 

Here is a screenshot of one of the subs. Notice how round the stars are. This is not under sampled and would not benefit from drizzle. At least not without a ton of data and not my much. If you inspect one of your subs, my bet is they are very comparable. Signs of under sampled would be square stars. 

 

Screenshot 2024-12-05 at 12.01.57 PM.jpg

 

Your fwhm is larger than I would expect at that image scale. I would look at your focus and perhaps guiding. I am less inclined to consider guiding given your good eccentricity. 

No I don't think they look square. It's true that I have slight problems with focus sometimes, I'm waiting for EAF, hopefully it will be better then.

I use OAG and ZWO ASI178mc, which I don't think is ideal for a guider, but the average RMS is around 0.7 - 1.5" (depending on the night).

Considering it's on Skywatcher AZ-GTI, it doesn't seem too bad to me.

[thomas7]

And you started with all the raw frames? Again, not a PI guy so not sure if it's possible - but it's easy to make mistakes in Siril. If I go back and try to drizzle already registered frames it gives me weirdness.

For stacking yes, then I used registered files that are ready for drizzle. I am also a beginner in PI, especially in drizzle processing.

[thomas7]

I can't point you to a specific thread on this without spending time searching this forum. My statement references dozens of discussions on this topic over the last several years paying attention to those advanced imagers that I trust. 

 

The concept is simple, you need intermediate data for this to work. That is you need data from 1/3 and 1/2 pixel over from the reference frame. This is done in a circular pattern around the reference frame. 

 

 

There are two reasons to dither. The primary one is move the fixed pattern noise around so it is eliminated by stacking. The second is to drizzle integrate 2x if your data supports it. 

So do you think Drizzle 1x would be better or is it necessary to make many more dither files in any case?

[psienide]

For stacking yes, then I used registered files that are ready for drizzle. I am also a beginner in PI, especially in drizzle processing.

If i'm understanding this correct - then I think that's the problem. Drizzle needs to be run against unregistered frames, otherwise it's not really doing anything.

[idclimber]

So do you think Drizzle 1x would be better or is it necessary to make many more dither files in any case?

I would not drizzle unless you have a lot more data and it shows signs of under sampling (square stars). Yes a focus motor will help.... a lot. 

[NMCN]

Seems like we need to work out what's the most efficient strategy for dithering with intent to drizzle. I don't seem to have any problems with dithering every 3 frames on 5 min captures. But if dithering every frame will afford some additional benefit I would switch.  

 

I have no doubt that I've taken my dithering to an unhealthy extreme. I doubt I'm getting any additional benefit by dithering that often and I plan to cut that back as I honestly don't recall how I got to those numbers anymore.

 

 

I can't point you to a specific thread on this without spending time searching this forum. My statement references dozens of discussions on this topic over the last several years paying attention to those advanced imagers that I trust. 

 

The concept is simple, you need intermediate data for this to work. That is you need data from 1/3 and 1/2 pixel over from the reference frame. This is done in a circular pattern around the reference frame. 

 

 

There are two reasons to dither. The primary one is move the fixed pattern noise around so it is eliminated by stacking. The second is to drizzle integrate 2x if your data supports it. 

Oh, I get the reasons for dithering. What I don't get is the notion of having a minimum number of unique locations. That was what I asking about. I disagree with your statement regarding a circular pattern, it should be random. The post I linked to was from 2017 and was more of a discussion regarding the amount of pixels each dither needed to cover to be effective, but Jon Rista was adamant that you do not want any kind of pattern in your dithering.

Edited by NMCN, 05 December 2024 - 03:21 PM.

[psienide]

For reference - this is what drizzling on data that doesn't really need or support it looks like.

 

1.3"/px

1.4x .35 drizzle for .997"/px

[idclimber]

I have no doubt that I've taken my dithering to an unhealthy extreme. I doubt I'm getting any additional benefit by dithering that often and I plan to cut that back as I honestly don't recall how I got to those numbers anymore.

 

 

Oh, I get the reasons for dithering. What I don't get is the notion of having a minimum number of unique locations. That was what I asking about. I disagree with your statement regarding a circular pattern, it should be random. The post I linked to was from 2017 and was more of a discussion regarding the amount of pixels each dither needed to cover to be effective, but Jon Rista was adamant that you do not want any kind of pattern in your dithering.

I highly respect Jon Rista on these issues. It is random, in both directions. You can dither on one axis, and is the only way with a single axis mount. This often results in the fixed pattern noise showing up in the integration. 

[NMCN]

I highly respect Jon Rista on these issues. It is random, in both directions. You can dither on one axis, and is the only way with a single axis mount. This often results in the fixed pattern noise showing up in the integration. 

Agreed. Seems we were just talking past each other.

[psienide]

Agreed. Seems we were just talking past each other.

I read it as describing the drizzling process as circular. When you look at individual drizzle registered frames you can see patterns that seem to support that.

[nhmorgan79]

Did you have WBPP debayer and already debayered image? I've done that by accident and it looked like that.