Deconvolve comparison

[kinetic (Steve)]

I've just had a play with Ken Crawford's positive
constraint Deconvolve tutorial on one of my recent
Star Cluster luminosity results. Open Cluster NGC 2516 in Carina.

Stopping short of doing the multi layer blending
which follows after the deconv/sharpen routine,
I have however improved the sharpness when compared to my
usual routine, which is nearly always too harsh.

1st image is my usual routine which is :

1. crop stack result (remove tracking errors on edges)
2. curves, (I usually use a Keller, QM or DS curve)
3. Sharpen
4. Final histogram adjust, usually set black
5. Noise reduction, usually edge preserving smooth.

2nd image is step 1. but then doing 30 iterations of Positive Constraint
Deconvolution in CCDStack before steps 2-5.

3rd image is a more aggressive 150 iterations of P.C. Deconvolve then
steps 2-5.

I have hardly ever used a deconvolution routine before sharpening on
Deep Sky stuff and am kicking myself! Quite excited though at the
prospect of repro-ing some of my better DSI II results.

Steve

[irwjager (Ivo)]

Call me crazy, but I think I actually like your old approach better! Personally, I find the deconvoluted images rather blurry and soft. But that's just me...

[kinetic (Steve)]

I know Ivo, it's a difficult thing for me to judge too!
Last night it was very obvious that the 3rd image looked far better,
crisper, better diffraction spikes, better blacks...smaller stars close
to the large brighter ones easily resolved. All on my good home CRT
and good house LCD.

Then I looked at the comparison on a work computer and it wasn't so
obvious!

Steve

[Alchemy (Clive)]

Interesting comparison...... But...... Have you recreated the processing steps exactly the same in each case, ie same amount of curves, same amount of levels etc ?

On the three pics provided, the last looks the best, the gap between the main star and the three or so upper stars is greater, but is this just due to less curves in the initial instance ?

My final consideration which probably is somewhat irrelevant in this case is just how a color image would handle this sort of processing.

[TrevorW]

I've found that deconvolution can often affect that background quality

[kinetic (Steve)]

Originally Posted by Alchemy

Interesting comparison...... But...... Have you recreated the processing steps exactly the same in each case, ie same amount of curves, same amount of levels etc ?

As close as I possibly could, Clive, yes.

I started with a crop for each but you can see one of the decon routines
actually creates a dim background disparity.

Steve

[RobF (Rob)]

I think the reality is its a fantastic tool to have in your kit bag, but you have to be prepared to work hard sometimes to use it to good effect, and ideally you want to be starting with plenty of high quality low noise, oversampled data.

Settings for faint wispy detail can sometimes be different to the stars, so might also need to selectively mask before applying too is another variable.

If you even get it all figured out can you give us all lessons Steve!

[bmitchell82 (Brendan)]

Deconvolution is a great thing, and works well in color images. be sure though to be gentle with it as it can be very image destroying. As Ken does point out that you have to do multi layered blended decon to make it work properly because it will make it look like the background has had sand on it.

keep it up

[Ken Crawford (KenC)]

Ok I will chime in here, like any sharpening tool you can use it like a surgeon's knife or a blacksmith's hammer!

To remove the "blur" from the data we get is not perfect but it can make a huge difference. Not all Decons are alike of course and some really add noise to the background, black ring stars, ect. I have found that I use 3 different sharpening methods on the same image.

(1) Multi Strength Decons blended with layers using Positive constraint for the high signal details. This is done in CCDStack2 Then brought into PSCS4 for blending. This way I can use a mild Decon to tighten the stars and the small faint fuzzies and a hard decon on for small details in the object and blend them together. This way I leave out the noise and the damaged stars.

(2) I will use different layers of High Pass filters with different blending modes so it works kind of like a wavelet filter. I use masks to blend in where I want it and to also stay away from the stars. I use a brush on the mask for precise control.

(3) I will then work on the tiny little details with a decon PS plugin called FocusMagic. This is used with a mask very carefully and selectively.

Then finishing off with blending modes like softlight and color dodge you can use color and contrast to make things pop and look sharper.

I am working hard on AIC2010 right now but will produce another tutorial to demonstrate putting all of that together if anyone would be interested.

Kindest Regards,

[RobF (Rob)]

Would love to see a tutorial on this sometime if you get get time Ken....!
(thanks for detail you've given soo far too...)