I want to give FocusMagic a thorough evaluation but 10 tests is not enough

[Bart_van_der_Wolf]

Hi Bart,

Thank you for your workflow on using FocusMagic.

Would you provide similar detail on where in the overall workflow to use FocusMagic? 

As I understand it, FM should be used before most everything else because of the nature of its algorithms.  So, my question is: what adjustments are safely used in LR before exporting  to PS where FM can be accessed?

Hi Doug,

Well, sharpening in LR would hurt the ability to use Deconvolution effectively. Some (color) noise reduction might be allowed if necessary. Local contrast adjustments do not help. Distortion correction / Upright do not help, because sub-pixel interpolation will create sharper pixels and softer pixels in alternating patterns of originally uniform structural detail. Chromatic Aberration correction does help.

I know that a while back there was a long thread on this matter, but it is not clear to  me what in summary was the best approach.

TIA for your contribution (again) on this matter.

You're welcome. The benefit of parametric editing becomes a bit of a drawback if Sharpening is not up to date. Especially the Deconvolution Capture sharpening should be applied early in the sequence of operations. LR could make good use of some serious improvements of it's sharpening functionality.

Cheers,
Bart

[Bart_van_der_Wolf]

Astra Image is very interesting, but it is very slow as it is single threaded and and no GPU... 

Hi Pictus,

I had a look and it look like a decent implementation of deconvolution, although I miss a control for noise regularizaton in the Deconvolution-sharpening section. Typical deconvolution may need some help with that, as part of the deconvolution process. The default algorithms they mention, do not have provisions for that, although they might have added some of that invisible to the user, under the hood.

The algorithms they offer seem to increase in robustness in the presence of noise, but that also depends on actual implementation. Noise free images (with very high Signal/Noise ratios) will do well with the Van Cittert algorithm, and Richardson-Lucy (especially if it has added noise dampening) is a well behaved general algorithm.

I do see different previews for the same parameter settings, depending on what one tries before, so that seems something that needs work. For example, I tried a deconvolution with a Blur kernel size (radius? not clear) of 1.00 which looked reasonable, then a deconvolution with a less critical radius of 0.99, which looked like rubbish, and then 1.00 did not change it back. So it's not entirely robust and bug free it seems. BTW, deconvolutions do use all 8 cores on my CPU.

It also produces ringing artifacts around the image borders, so one might need to crop a few pixels on all sides if that gets too obvious.

Cheers,
Bart

[DougJ]

Thanks for your reply to my post, Bart.

Doug

[Pictus]

Hi Pictus,
I do see different previews for the same parameter settings, depending on what one tries before, so that seems something that needs work. For example, I tried a deconvolution with a Blur kernel size (radius? not clear) of 1.00 which looked reasonable, then a deconvolution with a less critical radius of 0.99, which looked like rubbish, and then 1.00 did not change it back. So it's not entirely robust and bug free it seems. BTW, deconvolutions do use all 8 cores on my CPU.

Hi Bart,

4 cores here and only 25% utilization.

I saw some bugs too, do not remember where exactly, but some combinations
seems to change the highlights...
Maybe was in the Deconvolution Smart - Weighted Derivative + Blur Kernel Cauchy.

Lets wait for Astra Image 6. 

I will play more with Van Cittert/Richardson-Lucy and also investigate this Weighted Derivative...
Every time I read/hear the name Lucy this https://youtu.be/aVVjxPGDda4 comes to mind...