It appeared to work well on DP's data that I reprocessed, so I wanted to see what it could do for my data.
I've attached several images here:
- The first image has 3 versions. The top is the original, the middle is using VC deconvolution and the bottom LR deconvolution.
- The second image is a 1.4x resample in registax, and then using this method.
My conclusions? hmm I don't have any yet.
The resampled version is DEFINITELY better than my previous resample attempt on the same data. So maybe this method works best on large images?
The original is the best one there. It's got to a stage where you are trying to go beyond perfect and they are going backwards. There is only so much you can pull out of an image.
Stop beating yourself up about it, your original is fine!
Interesting how subjective this sort of thing is. The VC looks absolutely stunning to my eyes! The original is too "grainy" and artifically sharpened looking for me. LR is better but VC looks the best. At first glance it might look as if there is more detail in the original than in the VC, but if one picks out any detail in the original, it can also be found in the VC version.
The VC has a lot of subtle detail. The original has virtually no subtle detail! All the detail is very in your face. I'm guessing that's why it does not look as appealing and why the VC looks more natural.
How about doing one with ME (maximum entropy), Mike?
It's got to a stage where you are trying to go beyond perfect and they are going backwards. There is only so much you can pull out of an image.
Ken, I respectfully disagree.
Image processing (in my opinion) makes up 70% of the final result. Yes you have to use the right capture settings and a well collimated telescope at ambient temperature, but once you get the avi on a hard disk, there's a thousand different ways to process the same data.
And how you process that data can make a huge difference to the final result. If playing around with different processing methods can give me a 5% better image, then in most cases it's worth the extra effort.
If it means I post some images that "go backwards" while i'm testing out different methods, that's fine too. I've no problem with posting less than my best images, if it means:
a) I can find the best method to process my images
b) Others can learn from my experiences
Quote:
Originally Posted by ken
Stop beating yourself up about it, your original is fine!
I'm not beating myself up at all. I'm very happy with my original image. These images are simply an experiment, trial and error, to see if different processing methods can yield the SAME result, or a BETTER result.
Quote:
Originally Posted by steve
Interesting how subjective this sort of thing is.
Very true Steve. I guess you like the VC deconv image because it has a more "eyepiece-like" look about it, where as the others show greater contrast and sharpness than what you see in the eyepiece.
Quote:
Originally Posted by steve
How about doing one with ME (maximum entropy), Mike?
No problems, will do one soon and post it later.
Quote:
Originally Posted by dp
I am not sure with VC deconvolution, it leaves swirls like a car polish if you know what i mean. to me it is evident in the 1.4 resample
The 1.4x resample used LR, not VC. I agree though that VC can be very harsh. On the VC image I used default 5 and 2.
The "swirly" patterns in the 1.4x is (I think) due to lack of data in those areas, not enough frames stacked or not good enough seeing to increase the image scale.
Here's another comparison image, showing the original 3, plus 3 news ones.
The 3 new ones include:
- ME deconvolution (25/1.8) only. No other processing.
- ME deconvolution (25/1.8) + registax wavelets/gamma + contrast enhance + unsharp mask
- ME deconvolution (25/1.8) + registax wavelets/gamma only.
Thoughts?
I quite like the ME versions. Perhaps a tad smoother.
Remember to expand the attachment when your browser resizes it.
Of the MEs I like the non-unsharp-masked one best, but all the MEs have a mild case of the "onion" rings. Good chance this has to do with the "ringing" artifact often produced by ME and many other deconvolution methods.
You might like to try the ImageJ program. it's free and used by many professionals in scientific imaging. It allows manipulation of the image in Fourier (spatial frequency) space, amongst other things, giving you lots more flexibility than unsharp masking.
Mike,
Very useful thread. I personally like the ME results, though the
VC result is good as well. Curious about the original image.
Was there plenty of black space around the planet? I've found that
if you crop (say with ppmcentre or like tool) the image too tightly
then ringing around the limb is more likely. Also a PSF too large
can cause that as well. I think Bird is right about wavelets
discarding a lot of information. I think for purely cosmetic
reasons at the end of the image processing trail they are ok. The
amazing thing about ME is that if done properly it can actually
restore some original information!
Speaking of PSF, has anybody used the MaximDL ME deconvolution?
I've discovered that an exponential PSF can sometimes yield spectacular
results if chosen carefully (i.e. endless runs on the computer). When a
sweet spot is found the number of iterations is quite important as well.
I think the exponential PSF tends to correct for the diffraction profile of
our telescopes whereas the gaussian is better for correcting general
fuzziness due to seeing. In the real world we have both.
If one wants to get a good estimate at the PSF you can capture a mag 3
or so star (same config as your planetary stuff) careful not to saturate the
star image and of course careful with the focus. If you resample the
planet, resample the star as well. Then you have the closest thing to
the actual PSF, a good starting point anyway.
Spot on mike, the only way forward is keep practising, sometimes it is a backwards step, but in the long run, it is worth it. There are practical tips that can apply as "standard", but a heck of a lot of it is knowing what to do for a given set of conditions.
If one wants to get a good estimate at the PSF you can capture a mag 3or so star (same config as your planetary stuff) careful not to saturate the star image and of course careful with the focus. If you resample the planet, resample the star as well. Then you have the closest thing to the actual PSF, a good starting point anyway.
Thanks for sharing your results!
Glenn Jolly
Arizona, USA
Thanks Glen, sorry to be a dummy, but what do i do with the image once i get it.
do i try a few iterations until it is a pin point?
Once you have a good star image you can read it into MaximDL, just select on the PSF Model tab page "From Image" or alternatively if you save it in a fits file select "From File". I think internally MaximDL saves PSF images as FITS normalized to 1.0, but the original star image should not be saturated or you lose imformation. Hope this helps.
Its not an exact science, as sometimes I get a decent image, not especially different from previous ones, which simply will not yield to ME decon. Then I use LR decon - you know your done with LR when the Image Chi Square is minimized or bottoms out.
Sorry, your question was when is ME done? I wish I had some good info on that, I simply quit when it appears no improvement is being had. Sometimes it goes past a point where the image turns bad and does not recover so you have to remember how many iter back it was ok and repeat.
Many trials are made, and the result depends greatly on the dynamics of the image - the Red image is different than the Blue for example, and the number of iterations may not be the same - although you dont want to push it too far.
I do all my planetary imaging in RGB now, I've got the capture down, although I'm still using a manual filter wheel. But the RGB image processing is very time consuming. Tricolor is about 3^3 times more demanding of my time! Hopefully someday we will have better tools to do this!
By the way, when an image does not respond to ME sometimes a few iterations of LR before seems to help. But this is totally practical info - not based on any theory.
