Pixinsight Deconvolution is my new friend

[rmuhlack (Richard)]

Having a play this afternoon for the first time with Deconvolution on Pixinsight, using one of my early images from back in May. Still figuring out how the various settings impact on the final result (particularly settings for the star mask and deringing), but even so i think the results are simply amazing. Its like a fog has been lifted, and the detail that was there all along is finally revealed

The left image was processed using DSS followed by Darktable (a linux equivalent of Adobe Lightroom), the right image processed with PixInsight followed by Darktable.

The image itself is 54 x 4min guided subs @ ISO800 with a Nikkor 200mm f4 prime at f4 and a modded 400D

[multiweb (Marc)]

Gee the result looks dramatically improved. What steps did you do in PI?

[whzzz28 (Nathan)]

Looks great!

I must admit - i dont use deconvolution much in my images. I found it brought out noise. Lum filter helps though.

[Garbz (Chris)]

Nathan I agree with the noise but I have worked around it. I use a starmask or wavelets to extract all the stars for deringing, but then I also generate a second mask with both stars and bright nebula and then apply that to the image before deconvolution.

That way I keep PI from deconvoluting the noise in the dark sky.

[alistairsam]

Hi Richard
Just my personal preference, but I actually like the first one though it looks a bit out of focus. The star fields look more natural.
The one done in PI seems to have stars as perpendicular lines and looks a bit flat. The image posted is too small to make out if it is noise or other artefacts. Do you have links to higher res versions?
Stars in the first one aren't lines. Maybe there is another tool to be used after this??

[rmuhlack (Richard)]

Thanks for the feedback

Marc - I first created a star mask (on a linear image and before DBE), with the "aggregate" box checked and all other settings at default. Then using DynamicPSF i selected a bunch of stars with the Gaussian option checked and then used the "export a synthetic PSF" button. In the Deconvolution process window, I selected the synthetic PSF as a the external PSF, and checked Deringing with my star mask as the local ringing support, set at 0.5. I used regularised Richardson-Lucy with 15 iterations targeting RGB/K components (targeting luminance gave some very weird results). I then applied these deconvolution settings to my linear/preDBE image. Then process as per my current usual method (DBE, Histogram stretch, HDR Transform, Noise Reduction, and curves).

Alistair - I thought i'd use this image as the original looked to have bloated stars and/or was slightly out of focus, and so would possibly make a good candidate for some deconvolution experimentation. the perpendicular looking star patterns are simply an artefact of the resizing and compression applied for uploading the image to IIS - i dont see them in the full size images. My apparently less than optimal star mask and deringing settings are obvious in the full size image however, while the 'flatness' of the image i think can be adjusted with curves (my monitor might need adjusting also).



link to hi res image on astrobin: http://www.astrobin.com/21804/B/

[multiweb (Marc)]

Thanks for that. Never tried deconv in PI. I alwasy did it in CCD Stack. Will give it a go. Sounds like quite a process though.

[rcheshire (Rowland)]

Nicely done Richard. I must try that myself on a similar image.