Thoughts on LR Deconvelution applications.

[[1ponders] (Paul)]

I'm interested in thoughts, techniques and results of applying LR deconvolution processing to planetary images.

I felt the Robby made a very valid point about LR in Asimovs thread

I have always been higly suspect of doing LR Decon on planets and am of the view that the results are from good luck rather than true LRD. The reaon being is that with a stacked/processed planetary images there is essentially no information left in the image that accurately conveys the convolution of the optics. It has all been processed out. LRD is a process that theoretically should only be applied to RAW images before any processing has been done & the optical information is still present......

He does go on to say though

However that said, the consistant results that are shown here and in a lot of other planetary images (eg Mike's) indicate that benefit can still be had by using LRD on processed images. Perhaps a "standard best guess" PSF (point spread function) is used?

Also Asimov used the LR decon in his reprocesss without using any wavelets while from what I remember (and that is open to debate somedays) DP, iceman and bird us the LR after processing with wavelet (DP light, iceman hard9ish) bird? )


Any comments, thoughts or pontifications?

[davidpretorius]

I used to find LR was the one ie 3x1.3, but with latest image scale on Jupiter, a 3x1.3 ME works best, sometimes only a 2x1.1 ME on a heavily waveletted image worked well.

Back in decembers AS&T, there was an excellent article regarding how much info can you squeeze out of a system??

Using ME and LR on things like imaging detail on ganymede and IO along with the encke division.

http://www.stellarproducts.com/about...on%20tests.htm

The author of that article has this on the web, worth a read

[asimov (John)]

I've tried the same thing with the same AVI Paul as in my latest joop thread, but used the normal waveletts I would use. The results were not as good as my latest effort with no waveletts. I'll post the pic (if I can find it) when I get home from work tonight.

[davidpretorius]

First image, hard wavelets 79.8 on the 6th, 2nd image light wavelets, but more unsharp mask.

thes second also has increased contrast and decreased brightness.


both arrived at a similiar amount of detail

[asimov (John)]

My mistake. Mine with heavy waveletts & deconvolution was on a jpg. This was before I worked out I needed to save the registax final image as anything other than a jpg.

[Robert_T]

Hi Paul, I think the key comment of Robby's (as you point out below) on LR deconvolution is that it seems to deliver results even if it doesn't make sense why. I have little idea how or why any of the post-processing I use works, I just fiddle and see what delivers, and LR deconvolution delivers consistently improved results on most of my planetary avis. Not all mind you. I've found my best seeing avis are best straight from registax with a little soft-sharpening. Like DP the larger image scale shots seem to suffer too much noise with LR deconvolution with ME delivering better results. When I intend to run LR I still apply wavelets in Registax and export a BMP, but I take care not to push the wavelets too hard.

cheers,

[[1ponders] (Paul)]

So for good images a softly softly approach to wavelets before exporting. And when images are to be LR or ME processed in Astra Image you save them as BMPs, not 16 bit Tiffs?

[iceman (Mike)]

I save from registax as 16-bit TIFs, always.

[Robby]

Nice thread...
I have never actually tried LRD or any other D on planetary images. I know that it does help with my deep sky stuff sometimes.
I always apply the LRD to the Raws before stacking, and try to extract the PSF from a good star (or 2) within the image. However even with this approach I got mixed results so now I generally do not bother.
I was using the LRD within MaxImDL 4.
I think that the thoery is good but in practive there are many other factors (like seeing) that will kill your PSF. Any optical system will hava a known PSF that perhaps varies very little from session to session, so therefore one could generate a PSF on a night of exceptionally good seeing that can be used again... Pretty much thinking out loud here, so take it with a grain of salt!!!
Use what works & if LRD helps your image, then great!! If not, then hey at least you tried...
Keep it up!
Cheers

[davidpretorius]

ok,

http://www.precons.com/iceinspace/ga...r_060426_1147/

here are all the stacked tiffs from my recent "best" jupiter the file name will tell you what you are dealing with:

1.5_ is scale ie 1.5 mitchell resample

_ST99_ is the 99 best frames stacked

_WV_10_10_10_10_10_10 are the wavelet values from slider 1 to 6

I have no wavelets ie default = all 10's = 1.0 throught to heavy with slider 6 at 798 = 79.8.

please download and have a play with different deconvolution settings from VC, LR and ME and then maybe different unsharp masks.

is it possible to produce the same sharpness with no wavelets and heavier deconvolution versus heavy wavelets or does unsharp mask do a simliar thing.

What does damian peach do?

Any takers???

[[1ponders] (Paul)]

Ok I've done a bit of a trial here with an image from a moderately good night.

The first one is basically straight from registax with heavish 3,4,5 & 6 wavelets. The second is soft wavelets but saved as bmp. The third is soft wavelets but saved as Tiff. The third is hard wavelets and saved as Tiff.

Except for the first control the other three have been processed with 3 iterations of 1.1 in Astra image. All then (even control) were exported to PS CS and had levels and curves adjusted, then saved as jpegs.

I've tried to keep the processing the same for all AI images and no concession has been made for colour aesethic.

[Robby]

Nice work Paul. I guess it come down to exactly what you are after. In terms of detail #4 wins hands down. Tremendous detail round the edge.. Wow! But in terms of a pretty picture, #3 is more pleasing to the eye imho. And also #2 & #3 have moon detail, that is washed out in #4....
No clear winner here, but a great comparison.
Cheers

[Robby]

I just had a play with a couple of David's tiff with LRD in MaximDL.
I got moderate sharperning but could not use anything within the image to get a decent PSF. The only thing that would come close giving a reasonable PSF was the disc of the moon, but even that was "a bit rough"....
Here is what I got from one of David's "no wavelets" tiffs... Before & After.
Moderate sharpening, but not a successful LRD imho.
Cheers

[davidpretorius]

ok just adding the 10 best raw bmps to this directory as well

[StargazerX1]

David,

Here's what I got with ME. I used 9 interations with a PSF of 2.2
followed by some HF filtering and a mild unsharp mask. I also took some red out of the image.

I noticed the background had a lot excess blue (taken at sunrise?), and I also think if there was more black space around the image ME would do a better job since it would have a better noise sample.

I don't know why but some images will not work with ME. Yours did. Sometimes I get an image that simply disappears with ME and never comes back - even after 99 iterations! Its not science I guess. If the image will work with ME, then it will do a better job of controlling noise than LR.

If someone could tell us what logic to use in choosing a PSF it would be great. All I know is if the PSF is to large you get donuts for moons and nasty ringing around eclipse shadows and planets.

Glenn Jolly
Arizona USA

[davidpretorius]

nice processing glen, image taken at midnight and with optimized colour mode without a ir/uv filter.

yes, this technical area needs more investigation and discussion, not thread guys