ACDNR settings/star protection

[dhalliday]

Perhaps because I am using a new camera,my NR routines are not working so well....
Despite setting up what looks like a "clean" and adequate Lum mask,my background,and stars look "odd"...and the background can be "mottled"
I have played around with "bright/dark" side settings,and "star protection"...with some benefit,but not enough !!
Getting cleaner data is helping...,these are 12 minute exposures,at f/6.4; (5 each RGB)(I am being faithful to Juan's advice..)
http://www.flickr.com/photos/daveh56/4402286449/sizes/l/

Now here,in this rough process,I have used GreyC and Morphology...so stars look odd.
This is not my complaint though.Ignore the central star defects,and their pinched shape...
(Morphology settings are another story in itself !!)

In summary;regarding ACDNR;
1) what does changing the wavelet level in the mask do...??..just smooth it...?
       does this give more/less background NR..? (ie to empty sky..?)
2)or does it just do  what I see in the preview ?
3)Am I overstretching the image prior to ACDNR..? How "stretched" should it be...?
4) should I "hand make" my own Lum mask,and maybe apply NR to IT...??...and use that during ACDNR..?
5)what about star protection..? any comments on its settings ??
This of course comes back to the need for more comp. info in the parameter settings...
Any help appreciated...
Other (many !!) images of M1 are on my site...some showing this "mottling"...it is ugly as sin !!

Dave
PS The other "obvious" question is...what is bright/dark   doing..?
what is "overdrive" doing..?
when does one use multiscale,and recursive..?
what about SIZE...? default is 5...changing it does not seem to do much...

[mmirot]

The ACDNR processing did this to your stars? I have never seen it do that.

Max

[dhalliday]

Max
Sorry...I was a bit vague here...!
No- GreyCstoration gave them the bad cores...morphology shrank them,and bad tracking gave them an oval shape...( )
BUT...ACDNR gave them the wierd "halo" surroundings...and a "mottled" background...sort of a grey look...I  can only  darken the background adequately by clipping the left side of the data...then M1 half disappears...
I know I am not explaining this well...
In other words,I guess I am (mostly) asking about  choices with bright/dark side settings,and "star protection"...

Dave

[Simon Hicks]

Hi Dave,

I have had some mixed experience....so I'm no expert....but I would say the following;

1) I've had success getting oval stars more rounded by creating a starmask and then applying either....
       (a) deconvolution with a suitable aspect ratio and rotation, i.e. make it look the same as the stars or
       (b) morphological transform by painting a structure element that looks oval just like the stars.

2) The mottled background seems to be removed quite well by applying ACDNR in two stages. Firstly with a small Standard Deviation to remove the smaller scale variations, followed by a second application with a larger Standard Deviation. You will want to apply this through a mask so you are only smoothing the dark parts. And you might need to reduce the Bright Sides protection down very low.

3) Removing wavelet layers does exactly the same as opening up ATrousWavelets and placing a cross by layers 1, 2, etc. Basically its smooths at ever increasing scales. At least, that's what I think it does.

4) I usually hand make my own masks for ACDNR....you just have better control over them. And sometimes I apply smoothing or erosion or dilation to the mask depending on what you want.

5) Bright / Dark applies protection to edges or features in the image....you don't want to smooth these away or you loose features.  Overdrive is not usually needed....but you can use small values and see the effects....too much and it distorts the image.

That's me done!

Cheers
       Simon

[dhalliday]

Simon
Thanks...
Regarding star shapes,the more I look at Morphological trans.. the more I think it can "round out" stars by adjusting the crazy "way" thing...ie drawing a block picture of the bits of stars you want to round off...I am going to try that tonite...(forecast is hazy-darn !)
I will try the two stage NR...good idea..
Maybe I will post some more pics of my troubles tonite....bore everyone to death..

Dave

[Carlos Milovic]

Hi Dave

Just some tips, later I'll elaborate further:
- GraycStoration: Always use a luminance mask (inverted). It is very common that stars are destroyed by this process (remember that it was invented for daylight images). Also, since S/N ratio is proportional to the intensity, that masks acts as a very effective way to apply the denoising where it is more needed.
- ACDNR:
+ Use at least 3 iterations, with lower intensity values (near 70%, or less if more iterations).
+ Dark/Bright side protections work as a threshold to protect borders. If you imagine a border like a step, dark is the "downside" direction, while bright is the upwards direction. Lower values means more protection. This is the most crucial parameter, so fine tune it.
+ Overdrive is like a local contrast enhancement, something like a sharpening effect. For luminance is not needed. Chrominance may benefit from small values. 0 means no overdrive.
+ Leave luminance mask tweaks as advanced settings. Most times you won't need to change those settings.

That's for now. Simon already answered a lot of other things
+ Just in very noisy cases change the support for edges protection.

[dhalliday]

Carlos
Whew...!!!
I am getting a headache...
Just went back and applied the NR with dark side protection...and it seemed a LOT better...;
http://www.flickr.com/photos/daveh56/4404404891/sizes/o/

Without this,I was getting stuff like this;(or lesser variations of it...)
http://www.flickr.com/photos/daveh56/4405186116/

There seems to be a TON of things to pay attention to in ACDNR....
I will play around with it a bit more carefully...(!!!)

[Silvercup]

Hi Dave:

Simon and Carlos has answered your questions about ACDNR parameters. As Simon has said you can correct oval stars with deconvolution width the motion blur tab and a simple star mask. I have tried it with your image so you can see the parameters I have choose:



An this is the result comparation:



But if the stars has oval shape the whole image has oval motion so M1 too. What happen if we apply motion blur deconvolution with a luminance mask that has a brigth amount with a PIP exponential transformation:



We get this result:



Best. Silvercup


 

[dhalliday]

Silvercup
You sort of lost me a bit...but I WILL reread  a few times..!!
I DID just toss my image into "Decon" and actually used the same shape you did,to fix my stars..!!
I must be getting smarter !
http://www.flickr.com/photos/daveh56/4404482351/sizes/o/
But I forgot to use a star mask and so I guess I reDeconned the whole image...
M1 is getting a tad "fried" by the time I am done...
As far as the latter bit...PIP,etc...I will practice..!!
I will put the RGB TIF file (cropped) with dbe applied on sendspace and post the link...
I appreciate your time !

thanks!
Dave

[dhalliday]

PS....
Maybe you can tell me how you are able to put that screen shot up there...( )
Do you not have to post the image (screen shot) somewhere on the internet..?
(I sound like Al Gore....)

I keep having to put images online/on Flickr.. to show you guys what I am doing..
tedious..++
Is there a way around that...?
I feel like some kind of dummy here...

[dhalliday]

Here is the file
It is 5 frames (only) RGB,12 minutes
VC200L at f/6.4
SBIG ST2K...

http://www.sendspace.com/file/5x72dm

Maybe you can comment on my data...but be kind

Dave

[mmirot]

Dave Vista has a screen shot tool. Just go to help and do a search for screen shot
I believe you just paste the image in after that.

BTW, Silver Cup the deconvolution really improved the resolution 

Max

[dhalliday]

Silvercup
Leaving aside the complexity of PIP stuff,was your deconvolution applied to the image BEFORE any stretching,and NR,or after...?
I tried doing it on the UNSTRETCHED data last nite...and color preservation seemed MUCH better
Correct?
I am SLOWLY getting a bit sick of M1 ( ) but cannot move on until I get it better...!
Still hopeful to see what a Jedi might pull out of my raw stack...!!
I will look into "print screen"....just seems to "weird" for my Luddite brain

Dave

[Niall Saunders]

Dave,

If you want to post your images 'directly' into the Forum, you must still store them 'somewhere' on the Internet - your Flicker account should help, providing you can establish the URL of the image 'alone' - i.e. not the 'whole Flicker page', JUST the image.
I have just had a quick look at the way Flicker sets things up, and the URL is not immediately accessible. If, however, you click on the little magnifying glass <All Sizes> icon, this takes you to a page where you CAN get the URL for the image alone. On the next page that appears, there was a banner that says <Download the Original Size>. Right-click on that, and then select <Copy Shortcut>. This will store the URL in the clipboard.

Then (back in the PixInsight Forum <Reply>, or <Create> window) you need to click on the <Insert Image> icon above (the second in from the left, on the second row, underneath the <I> icon). You simply enter the URL into the 'space between' the two sets of brackets - et, voila! (hopefully).

Double-check that the link does what you want it to do by clicking the <Preview> button at the foot of your 'reply', your image should appear in the preview.

Here is a test, based on using       http://farm5.static.flickr.com/4036/4405186116_15750e44e9_o_d.jpg      as the data 'copied' from one of your (random) Flicker pages:-



Cheers,

[dhalliday]

I was distracted by the Emoticons....
will pay more attention to the buttons..

But I will still have to "print screen" (from Pixinsight"...) and put it ON Flickr...?
In other words,your telling me a faster way to copy/paste the flickr image..(and I appreciate it !)...but it has to "go through flickr"..?

Dave (the dinosaur  8))