M27 - Deconvolution Issues

[niteman1946]

Looks like the day for decon problems.  The attached screenshot shows the following:
1.  Upper right - image prior to decon procedure
2.  Upper left - image after decon with "global dark" at 0.13
3.  Lower left - image after decon with "global dark" at 0.25
Used the PSF, appied a light mask and cited the star mask in the decon routine.

Pushing the global dark to higher levels reduces the halos around the stars in the nebula, but 0.13 is as high as I can go before the background starts to look like the lower left image.

The star mask does not pick up any of the stars in the nebula, only the surrounding areas.

The halos (wringing?) have become more problematic as more photons are captured and the nebula has become denser.

All suggestions are welcome.

http://www.astrobin.com/full/23767/?mod=none

Mark

[sleshin]

In my experience, to control ringing you always need to lower the default  Global Dark setting. Try 0.03 or 0.04.

Steve

[Philip de Louraille]

Are you applying this filter on a linear image or is it past the Histogram Transform?

[niteman1946]

Thanks Steve.

The lower values lead to more prominent ringing - I started at zero and worked up.  In this case increasing the gd to 0.130 mitgated some, but not enough.

I also tried Manuel's approach (Parametric PSF), but the stars came out "pinched" and eccentric.
http://www.manuelj.com/Tutorials/Deconvolution/22071685_r3Z6QC#!i=1767553866&k=NKhtp5J


Thanks Philip,
All images are linear.

Mark

[marekc]

Hi Mark,

Although I don't have anything particularly useful to add just at the moment, I send sympathy! I have been struggling with the deconvolution of an M33 luminance image for about a month now. It is just one problem after another!

Dark ringing artifacts have been one of my biggest problems, particularly around stars that are seen against the brightest part of M33.

I have been endlessly iterating parameters with the VaryParams script, and am just now starting to get something half-decent. In a perfect world, maybe I'll wind up with some useful insights after I'm done.

Here's something I've been wondering about - and I might make a separate thread about this... Can we re-linearize an image? I'm thinking about this for star masking. As I think you've noted, the default StarMask settings don't catch all of the stars that are seen in front of the bright part of our object. I'm wondering if it's worth doing something like this, in order to make the local deringing mask:

1) Clone our image while it's still at a linear stage.

2) Take it non-linear with a Histogram Transformation, such as by applying AutoSTF to HT.

3) Use HDRMT to reduce the dynamic range in our DSO, so that the object (galaxy, nebula, whatever), is mostly gone, or at least largely grayed-out.

4) Take this non-linear image, and somehow UNDO the HT from step 2, so we're back in linear-land, albeit with the DSO mostly gone.

5) Apply something like a default StarMask routine to the `re-linearized' image, and hopefully this time we'll catch the stars that are in front of the galaxy. This might make a better local deringing support.


I guess I'm envisioning `going back to a linear state'. Maybe this is about as realistic as regaining one's... er, never mind     , but it might be useful if it could be done. Just a crazy idea I've been thinking about.

- Marek

[pfile]

maybe try something along the lines of what's in these process icons? they probably won't work straight away because you have to tune the values in the stretch and pixelmath, and also the threshold for the starmask. but you should be able to tweak it into shape.

also the process icons are at the top of workspace #2 in case you cant find them...

[marekc]

Thanks very much for those process icons, I'll give them a try (with, presumably, some parameter tweaking) once I figure out how to open them. My Mac downloads them as files that end in .xpsm.xml. If I can figure out how to get them to just be Pixinsight project files, I'll hopefully be able to check out the process icons. I'll fiddle with the downloaded files some more, and hopefully I'll have some luck.

[pfile]

that is weird - i uploaded them from a mac. probably if you just delete the .xml portion and leave .xpsm it will work okay.

the idea is pretty much what you outlined above, but after the HDRWT i used pixelmath to delete all the dim stuff, then removed some of the small-scale stuff and then ran the StarMask process on that image. i think you might be able to get similar results by messing with the shadows/midtones clipping sliders in StarMask but i've always had trouble with that since you have to do it 'blind'. it would be nice if you could get a RT preview of the clipped image.

[lucchett]

Hi guys,
I normally use HT to replicate the shadow/midtones adjust in starmask, but here you have the option of the realtime view, and can be run also on a preview. the adjustment is quick

[niteman1946]

Marek,
Yeh, I spent a whole day trying to get something in decon to work and finally decided to throw the problem onto the forum.  Sorry you're having so much trouble too.  And thanks for the suggestions.

Pfile,
I opened your process icons and just used your default settings as is (after I found them on ws#2).  It made a nice star mask and picked up most of the stars in the nebula.  I did not use your mask, but should have since it would have brought more to the table.

Lucchett,
Thanks for the input.

Lacking context I suspected the absence of starmask stars in the nebula was the main problem.  Now I don't think so, although as I said before, pfile's mask would have given more flexibility in finishing the image.

What frustated me about Juan's "psf" method was that it was too aggressive and I had no way to tone it down.
But following everone's input, I found that by adding a slight value to global dark in the external PSF tab (0.005 in my case), and by reducing std deviation in the parametric PSF tab (from default 2.00 to 1.75), I got sharpening without ringing.

It would have been nice if someone had said "here's your problem, and here's your fix" .  But the oblique method seems to have done the trick .  FWIW I did upload to endor the  image prior to the decon procedure if anyone would like to take a shot at it.  Just let me know how to point you to it.

Thanks again guys,

Mark

[Alejandro Tombolini]

FWIW I did upload to endor the  image prior to the decon procedure if anyone would like to take a shot at it.  Just let me know how to point you to it.

Hi Mark, you have to upload it in the Forum Shared File section to allow us to download it. 

Regards, Alejandro.

[niteman1946]

Thanks Alejandro,

Will do.

Mark

[niteman1946]

Hi Alejandro,

The fit file is located in Forum Shared Files under
niteman1946/Deconvolution M27

Mark

[Alejandro Tombolini]

Hi Mark, is this image the result of the integration or it has any process done?

There are some strange shapes in the background. I have exagerated the stretching to see it. This makes difficult the use of deconvolution.



First generate the PSF to be used in deconvolution:



This are the settings used for deconvolution. You can also generate a starmask and use it to protect the cores of the biggest stars. 



Regards, Alejandro.

[niteman1946]

Hi Alejandro,
Thanks for taking your time on this image. 
a)  The fit file represents about 21 hours of integration at -10C with an Atik 383L+mono with 10min exposures.  This was done on a 12"LX200 Classic with F6.3 reducer, off axis guider and Orion skyglow filter.  Used Bias, Flat and Dark Masters all at -10C.  Flat and Dark Masters were not calibrated, but only integrated.  Calibration of Flat and Dark was done at Light calibration.
b)  The fit file was cropped, dynamic background extracted, and then multiscale median transformed for noise reduction.
c)  Crud in the background .... looks like remnants that MMT noise reduction could not clean up.  I can send the integrated file with no process if you would prefer.

So many questions ......
1.  Should decon be executed before any of the steps in the above "b"?
2.  Did you apply a luminace mask to the image prior to decon?  You don't mention it.
3.  You did not use a star mask in this example?
4.  You did not modify any values in Parametric PSF?
5.  What drove your decisions for the values in Deringing (especially global bright)?
6.  What drove your decisions for the values in Wavelet Regularization?

Very appreciative,

Mark