9 replies to this topic

[dcaponeii]

David Biggs sent me the latest version of BiggSky for testing over night (beta_v0_2_15) and having seen Mike's Jupiter processed using BiggSky the other day I thought I'd give it a go on some of the data I've been using to conduct final testing of Wavesharp 2.0.

In each of the panels below (full 1600px width can be accessed by selecting the image set) the LEFT hand image is processed using BiggSky Blind Deconvolution with the default settings with the exception of adjustments to the psf radius and the Guard Band settings. The default settings for both these parameters are 32. The comparison images are processed entirely in Wavesharp. The center image is what I consider "nominal" sharpening. Layer 1 = Strength 0.150 and 100%, Layer 2 = Strength 0.130 and 40%, Layer 3 = Strength 0.110 20%. The right hand image is what I consider to be "aggressive" sharpening. Layer 1 = Strength 0.180 and 100%, Layer 2 = 0.150 and 40%, Layer 3 = 0.120 and 20%. FFT Denoise and RGB Balance along with a 20% Saturation bump have also been applied to the images (including the BiggSky processed images). For the two images processed with psf radii smaller than the default value of 32 I have applied wavelets to the image following deconvolution. The values used in those two cases are presented with the image set below. Finally the comparison composite images were assembled in Paint Shop Pro in which curves and white balance were applied to ALL of the images.

First set is BiggSky Blind Deconvolution with psf = 8 and Guard band = 8. This image was sharpened in Wavesharp using Layer 1 = 0.100 and 100%, Layer 2 = 0.100 and 40% and Layer 3 = 0.100 and 20%.




The next set is psf = 16 and Guard band = 16. This image was also sharpened in Wavesharp using just Layer 1 = 0.100 and 100%




This image is psf = 24 and Guard band = 24





This image using the default settings of psf = 32 and Guard band = 32. No additional sharpening was performed on the image.





This image using psf = 64 and Guard band = 32 Previous work on earlier data convinced me that a Guard band larger than 32 was unnecessary.





Finally, this last image uses psf = 128 and Guard band = 32.




In summary, my conclusion is that BiggSky in the current version is a good option for sharpening Jupiter images. The default settings of psf = 32 and Guard band = 32 yield acceptable results. Increasing the psf radius quickly begins to yield less satisfactory results and I think there is some value in combining blind deconvolution with wavelet sharpening instead of relying solely on blind deconvolution for sharpening.

I'm going to spend some time working with some Saturn data next as Con gets some spectacular results with his data using deconvolution as an alternative to wavelets (he uses Astra Image if I recall correctly) and I'm curious if blind deconvolution can be effective at bringing out finer details in the rings than wavelets does alone on my data.

Edited by dcaponeii, 03 December 2024 - 03:48 PM.

[Wdh]

David Biggs sent me the latest version of BiggSky for testing over night (beta_v0_2_15) and having seen Mike's Jupiter processed using BiggSky the other day I thought I'd give it a go on some of the data I've been using to conduct final testing of Wavesharp 2.0.

 

In each of the panels below (full 1600px width can be accessed by selecting the image set) the right hand image is processed using BiggSky Blind Deconvolution with the default settings with the exception of adjustments to the psf radius and the Guard Band settings.  The default settings for both these parameters are 32.  The comparison images are processed entirely in Wavesharp.  The center image is what I consider "nominal" sharpening.  Layer 1 = Strength 0.150 and 100%, Layer 2 = Strength 0.130 and 40%, Layer 3 = Strength 0.110 20%.  The right hand image is what I consider to be "aggressive" sharpening.  Layer 1 = Strength 0.180 and 100%, Layer 2 = 0.150 and 40%, Layer 3 = 0.120 and 20%.  FFT Denoise and RGB Balance along with a 20% Saturation bump have also been applied to the images (including the BiggSky processed images).  For the two images processed with psf radii smaller than the default value of 32 I have applied wavelets to the image following deconvolution.  The values used in those two cases are presented with the image set below.  Finally the comparison composite images were assembled in Paint Shop Pro in which curves and white balance were applied to ALL of the images.

 

First set is BiggSky Blind Deconvolution with psf = 8 and Guard band = 8.  This image was sharpened in Wavesharp using Layer 1 = 0.100 and 100%, Layer 2 = 0.100 and 40% and Layer 3 = 0.100 and 20%.

 

 

 

The next set is psf = 16 and Guard band = 16.  This image was also sharpened in Wavesharp using just Layer 1 = 0.100 and 100%

 

 

 

This image is psf = 24 and Guard band = 24

 

 

 

This image using the default settings of psf = 32 and Guard band = 32.  No additional sharpening was performed on the image.

 

 

 

This image using psf = 64 and Guard band = 32  Previous work on earlier data convinced me that a Guard band larger than 32 was unnecessary.

 

 

 

Finally, this last image uses psf = 128 and Guard band = 32.

 

 

 

In summary, my conclusion is that BiggSky in the current version is a good option for sharpening Jupiter images.  The default settings of psf = 32 and Guard band = 32 yield acceptable results.  Increasing the psf radius quickly begins to yield less satisfactory results and I think there is some value in combining blind deconvolution with wavelet sharpening instead of relying solely on blind deconvolution for sharpening.

 

I'm going to spend some time working with some Saturn data next as Con gets some spectacular results with his data using deconvolution as an alternative to wavelets (he uses Astra Image if I recall correctly) and I'm curious if blind deconvolution can be effective at bringing out finer details in the rings than wavelets does alone on my data.

Watching this with interest, Don!

[Borodog]

Don,

 

What were your denoise settings in BiggSky? I find the post-denoising too aggressive, leading to artifacts. I typically leave post-denoising for waveSharp, which as you know is excellent. However the pre-denoise, which denoises the image for the PSF estimation only, seems usually produces good results. I've been using it at the default value of 80%, but have not experimented with it in detail.

 

Another thing to consider is the raw blend feature. If the deconvolution starts to become too aggressive, bringing up to much noise or resulting in excess contrast, BiggSky allows you to blend back in some fraction of the raw image to tame these effects.

[Borodog]

Also, I can't tell from your description which image is which.

 

In each of the panels below (full 1600px width can be accessed by selecting the image set) the right hand image is processed using BiggSky Blind Deconvolution with the default settings with the exception of adjustments to the psf radius and the Guard Band settings.  The default settings for both these parameters are 32.  The comparison images are processed entirely in Wavesharp.  The center image is what I consider "nominal" sharpening.  Layer 1 = Strength 0.150 and 100%, Layer 2 = Strength 0.130 and 40%, Layer 3 = Strength 0.110 20%.  The right hand image is what I consider to be "aggressive" sharpening.  Layer 1 = Strength 0.180 and 100%, Layer 2 = 0.150 and 40%, Layer 3 = 0.120 and 20%.

[dcaponeii]

Don,

What were your denoise settings in BiggSky? I find the post-denoising too aggressive, leading to artifacts. I typically leave post-denoising for waveSharp, which as you know is excellent. However the pre-denoise, which denoises the image for the PSF estimation only, seems usually produces good results. I've been using it at the default value of 80%, but have not experimented with it in detail.

Another thing to consider is the raw blend feature. If the deconvolution starts to become too aggressive, bringing up to much noise or resulting in excess contrast, BiggSky allows you to blend back in some fraction of the raw image to tame these effects.

Not a fan of either of the BiggSky denoise options so I do all my denoise in WS. I understand about blending but my point was to gauge the ability of the blind deconvolution itself.

[dcaponeii]

Also, I can't tell from your description which image is which.

Duh!! Fixed it in the post. The LEFT hand image is BiggSky. However, the fact that it’s hard to tell until you start using really large psf radius is a good result I think. Between psf=24 and psf=64 there’s some distinct advantages to the blind deconvolution.

[Borodog]

Not a fan of either of the BiggSky denoise options so I do all my denoise in WS. I understand about blending but my point was to gauge the ability of the blind deconvolution itself.

Give the "Before" denoise another look sometime. All it's doing is denoising to get a cleaner PSF.

[dcaponeii]

Give the "Before" denoise another look sometime. All it's doing is denoising to get a cleaner PSF.

I used it during this sequence but it really makes things "blotchy" (like someone brush-painted the fine details) so I left those trials out of the comparisons.  It actually appears to be the thing that I noticed in your image the other day that I described as "aggressive".  80% Prenoise works well on the Lunar images as long as the background isn't too bright (then you get the same effect) but for planets I just don't like the outcome at all.  FFT denoise is so much cleaner and leaves the details intact.

[dcaponeii]

Don,

Here's my first image from last night, sharpened in BiggSky with a tiny smidge more in waveSharp. 21 minutes derotated after BiggSky, processing finished in waveSharp.

2024-12-05-0210_6-MPO-Jupiter-derotated_WS_8bit.png

I saw it in your post. Mother said if you don’t have something nice to say…so I didn’t comment there but there is something seriously off with these recent images. They almost look like image doubling with a really close separation. Also the rind caused by WJ is pretty harsh too. I wish I could put my finger on it but I don’t know the cause. Plenty of good images at your scope size. Even my stuff with the 10” from 2 years ago is sharper details than these recent images. Maybe you’re pushing the sharpening too hard?

[dcaponeii]

Not seeing it. 

Not a problem.  Took me two years to finally understand what Daryl was telling me all the time about my early images too.  Make some side by side comparisons for yourself with other imagers using the same class scope and I think you'll start to see what I mean.  If not it's not a concern as they are your images and always that take precedent over what anyone else thinks.