Composite is interesting. Depending on how that Synth L is blended together, and depending on what the star is emitting, I suppose that mashup could potentially cause PSF pertubations in any multi-channel data.
A side by side of the same screen and settings between 557 and 558 does seem to show the new version tames down those severe black eye rings on certain of the stars.
They are still there though, but at least more within the deringing parameters capability to handle. Linearity may still need to be dropped to a very low setting.
The other matters I think still exist in testing, so I'm not sure we're totally there yet.
I used Stefan's M8 data, since that seems to present the major snags in one nice package, and ran it through 558, 536, 557, and then 558 again. Same steps each time. Composed as bicolor OSC, crop two mouse wheel clicks on each side, bin 50, wipe default, optidev no-ROI IFD-5.0, contrast default, and then into SVD.
I am still seeing a big dichotomy between stars in the SVD selection mask, and those without. This is also (barelyt) noticeable in synthetic deconvolution too, if one doesn't start selecting samples right away and lets it finish. Non-SVD-mask stars may end up with black eyes, albeit lesser now in 558, while SVD-mask stars don't seem to. They also act differently to the deringing controls.
Turning dering fuzz completely OFF seems to reveal some shapes that may be a deringing support mask. On a wild guess, maybe something like that is created by taking the existing SVD mask and adding to it via Mask procedures. But, there could be some flaws in that creation - again, guessing? Some of the deringing shapes (no fuzz) start out very blocky or even square, like what can happen with circles in Mask after using grow, though they still seem to need at least another hit of grow, and more round would be good too. Heavily lowering linearity can help round those off a bit. I also noticed several "misses" in those deringing square blocks in Stefan's data near the core -- one was displaced upward from the star's profile itself (leaving the bottom half still ringed, even after fuzz was added back in), and as a second I found some bright non-star nebulosity was included, causing structural detail be fuzzed out, after SVD had so nicely revealed it.
All of that strikes me as very similar to an AutoMask issue, with the starfishy SVD selection mask stars doing the best, and the non-starfishy added masking not so much. That may also explain why similar stars end up with different results, including tiny stars against a nebulous field that don't get any deringing effect at all.
Of course, I have no idea if I am in the ballpark or just completely loony.
And sorry for that screen-filling screenshot. Not very discreet ...
