When to Deconvolve

[dcalleja]

Hi
I've been playing with some of the features in CCDStack and tried deconvolving an image of the Tarantula I took a couple of weeks back. Heres a comparison image (I tried to keep the CS3 process flow the same). My question is; when should you use this?

I think the deconvolved image looks better (tighter stars) though I had to play around to avoid a doughnut\halo effect around the stars.

[multiweb]

I think deconvolution works well when your image scale is around or under 1asp and your stars are "fat". CCDStack Devonvolution algo is great [most of the time]. Recently though I have been deconvolving (very slightly) wide field pictures with already small stars. I have found that deconvolution on an already recombined color image can potentially damage the stars in the blue channel in particular for my QHY8. You get stars with a hole in the middle. The latest version of CCD Stack (v2) tends to do that on a color 32bits FITS file. Never had this problem before. Just found out because I had some troubles getting blue colors through in my latest shots with the ED80 and when looking at the blue channel in Photoshop most of the stars had litterally been punched out. Big black hole in the middle leaving only a ring but the green and red were just fine. No wonder I couldn't get blue through. :lol: So my advice if you're going to deconvolve your pics is to do it prior to color recombine and try to match your star sizes in all three channels.

PS: I've just done a bit more testing and found out that using the Saturation in the Color Dialog box is what create most of those holes. So deconvolution is not the culprit.

[dcalleja]

Marc
I definitely had fat(ter) stars in the original image (poor focus I suspect?). So I think the final deconvolved image was much improved. My cam is OSC colour so not sure how the colour channels are affected for me. I do the deconvolve on the final registered debayered image. so maybe I should do it on the RAWS

[gregbradley]

Also deconvolution as a tool needs to be managed well. It can very easily be overdone. I use it sometimes on the luminance and even on the rgb but decided recently not to do that anymore as it takes away the soft outer halo of the stars which is one thing that makes APO images so appealling.

Greg.

[dcalleja]

Thanks Greg. I'll take a look at few different image types to compare. I don't think I had soft outer halos on my stars though, I suspect they were just chubby

[TrevorW]

Never use it on my images

[Hagar]

Looking at the times I have used it, it seems to be an as needed tool which can easily make or break an image. The skill in deconvolution seems to be in it's use, to use or not to use and then how much to use. Certainly it can help to reduce stars which are oversampled but can also destroy them just as easily. Ken Crawford has a great tutorial on using deconvolution on his site which does give some aspects of it's use. I have just purchased Adam Blocks DVD tutorials on CCDStack and I will put up a report on it when I get time to have a look at it. (9Hrs of video).

[RobF]

That's where you need to use the extended features in PixInsight's deconv routine like deringing and wavelets/noise :evil:

It does seem to be a black art though. From what I've read and meagre experience plenty of detail (oversampling) and lots of data is really required to get maximum benefit, otherwise you run the risk of highlighting noise in the subs.

[Bassnut]

Decon is just magic if used appropriately, what number of itterations did you use?.

It suits narrow field/long FLs as Marc and Rob says at 1 a/s or more and oversampling. It reduces star size very effectively, but where it shines is extended nebula where itterations of 200 or more are common with Ha data and selective application with PS masking. IMO it is the most powerfull filter tool available for astro images if used selectively.

[marc4darkskies]

Decon is one of the most powerful tools in your processing toolkit. It's unforgiving of poor data though so only apply it (via layer masking) to the higher S/N areas of your image. Also, don't use it as a sledge hammer. Use it gently and with finesse.

Cheers, Marcus

[drjcaron]

Hi All,

There are all kinds of little-used image metrics to determine whether an image has blur or not. However, our eyes are the best indicator or blur, since they have spent most of a lifetime focusing to remove blur from our vision. Thus, if an image looks to have some blur, you should apply a deconvolution to try to remove it. The success depends on whether the blur through the optics has been retained in the image data, and whether your (blind) deconvolution method has extracted the point spread function successfully.

A couple suggestions, 1) apply the deconvolution after stacking but before any other operations, 2) do not use any image compression, 3) save the averaged frame from the stack in 32-bit single format. These things will help preserve the blur information in the image and produce a better deconvolution.

Hope that helps,
Jim Caron

[Bassnut]

Jim

Yes, I have found the points you make valid just by trial and error ;-).
I just had a look at your product (http://www.quarktet.com/index.html), looks very interesting, I wasnt aware of it, im tempted to try it, many of us here use CCDstacks decon filter.

Out of curiosity, you use blind decon. Is it possible to caracterise a given telescope optics and seeing to produce an umm, "not blind" algorithm? (thats a stab in the dark, a blind guess ;-).

[gregbradley]

How do you do that Marcus? Don't you apply it to the master stacked luminance? Or do you save both the stacked lum and deconvoluted lum and open in photoshop and apply a mask to let only parts of the decon through in the detailed part of an image?

Not a bad technique -no wonder your images are so good - you are very tricky!

Greg.

[Ken Crawford]

I have found that Decon can help almost any type of image, if used selectively. There are many decon methods out there but I have found the best way for me is to use different strengths and blend the best of the results together. One mild strength to tighten the non-saturated stars and a strong one to bring out the details like in a galaxy core or dust lane. Then using layer masks with normalized streatched frames blend in the details without draging in the noise or the messed up stars.

I call it Muti Strength Decon Layer Blend and you can see a basic step by step here.

http://www.rdelsol.net/ProcessingToolBox/ProcessingTool%20Box.html

There is also a write-up in the May S&T on it.

Kindest Regards,

Ken Crawford

[RobF]

Funny the way things pop up when you're ready for them. I'm sure I've visited your site before Ken and most of this tut blew over my head. Just 10mins ago I was digesting a 2008 Astrophotoinsight with a reference to your webpage and your multi strength deconv thinking what a great idea, and now here you've popped up on IIS re the same article.

Great work too by the way - also looking forward to getting my head around your techniques for enhanced contrast/3d effect :thumbsup:

[dcalleja]

Fred
50 iterations. It took a few trials to work out what was best. At 100 I ended up with doughnuts

[drjcaron]

Hi Fred,

The answer is yes. The easiest method is to image a star to serve as a point source. This is a standard approach with observatories and was used with the Hubble before the optical correction in 1994. It has subtle pitfalls. For example, the size and shape of your blur function (called a point spread function (PSF)) depends on the color of the object. If you use the star to clean up an image of Mars, you may get ringing effects near sharp edges.

You can also model the telescope and atmosphere to create a synthetic PSF. I have never heard of this actually being done. More likely, one would create a synthetic PSF and use it as a starting point for a blind deconvolution method.

Best Regards,
Jim C

[Ken Crawford]

I agree that finding the best PSF is great but I found that getting close if good for me.

With galaxies it is much easier as you can hit the galaxy very hard as it's signal is very high.

So I normally do at least 100- 150 iterations with CCDstacks positive constrant. It rings the saturatued stars but I do not use them as I blend in the decon details of the galaxy into orginal with a hide all layer mask. After "painting" in the mask to reveal the galaxy details I then feather the mask with a blur or with PSCS4 you can directly feather or adjust the mask edge.

Then boom - detailed galaxy while leaving the stars alone.

You can use a small (20 -50 iterations) decon to tighten the smaller non-saturated stars then use a mask to blend in the larger stars.

Then you end up with a super sharp Lumiance to add to your RGB. Once the RGB has been added and adjusted then I use a PS plugin called FocusMagic - but that is another story.

Regards,

[Bassnut]

Of course, I get it now, thats what CCDstack does, doh :P.