ST Modules and Scale

[Mike in Rancho]

Oops. You are of course right, Ivo. It's an illusion!

I ran the same data side-by-side and AutoDev stretch - weak or strong, results in the same underlying color cues. Likewise with the binning percentage. But a different bin, or a different crop effectively causing a different relative bin, will alter the quality colors. SNR stuff, no doubt.

The AutoDev can, though, affect the white outlines significantly, as to both shape and size of the profile detected, and the fade away from core. And even sensitive mask won't fully make up for a "better" initial stretch with regular apod mask.

Funny how a beautiful big round star one way, that I would totally jump on with a blue box, can have square sides another way, such that I probably wouldn't select it.

My few times using AltStars, no matter what combination of settings I change, just gives me a sea of starfishies. Or so it appears. Same again trying it just now in SVD -- I can't seem to get separation and in fact AltStars seems way more prone to capturing not only close doubles, but all its other friends within a few parsecs. The white outlines look like a crazy jigsaw puzzle though!

[admin]

The AutoDev can, though, affect the white outlines significantly, as to both shape and size of the profile detected, and the fade away from core. And even sensitive mask won't fully make up for a "better" initial stretch with regular apod mask.

You are right. This is one of the ways deconvolution in ST differs from legacy implementations. Here ST is taking into account how the image was stretched and how much ringing (in particular) is likely going to be visible. In the next version of StarTools masks are no longer used for deringing at all (this has been long on my wishlist - the less subjectivity for these sorts of matters, the better). Testing so far seems pretty solid.

The latter use for a star mask not withstanding (for deringing), star mask genetration is not exactly what SV Decon needs. Instead it needs something much closer to what AltMask generates; we're not interested so much the human visible concept of a star, but we are much more interested in the mathematical concept of a point spread function. This bring is us to;

My few times using AltStars, no matter what combination of settings I change, just gives me a sea of starfishies. Or so it appears. Same again trying it just now in SVD -- I can't seem to get separation and in fact AltStars seems way more prone to capturing not only close doubles, but all its other friends within a few parsecs. The white outlines look like a crazy jigsaw puzzle though!

During testing, I have found the AltStars mask generation generally superior in three ways;

1. It actually provides a good, tapering-off PSF, even if it is "polluted" by non-usable pixels. Whatever can be used, is used. If gaps exists, it they are filled in by other samples that may/will have those pixels covered. (indeed, this can yield starfishies depending on your dataset and scope's/optics characteristics). This in turn yields more accurate samples and slightly improved end results.
2. For elongated samples (coma, tracking error), it tends to better detect/model these sorts of PSFs.
3. I have found it much easier to find good candidates, scanning around the image.

The only drawback is that the AltStars mask generation algorithm does not include diffraction spikes. Inclusion of diffraction spikes can definitely help improve decon results. This is the reason why, in the next version of StarTools, you will still find the "old" star mask generation as an option for creating and apodization mask.

[jhart]

Hi Ivo,

Are you saying that if AltStars gives you green starfish on your stars (or in my case squashed starfish with like 15 or 20 little arms) that you should go ahead and keep that mask and proceed with Decon?

Jeff

[admin]

Are you saying that if AltStars gives you green starfish on your stars (or in my case squashed starfish with like 15 or 20 little arms) that you should go ahead and keep that mask and proceed with Decon?

That's correct. Same rules still apply of course; don't include sample boxes that have multiple stars in them, nor part of other stars.

[Mike in Rancho]

Alas, I must admit defeat in my attempts to use starfish for SVD sampling.

The theory seems sound, but in practice, the starfish are just too interconnected to get separation for sampling. It would take a lot of switching back to Mask, but I suppose I could use single-pixel-off to carve a ditch between merging profiles (requires some guesswork) and then erase the next-door star(s).

I have tried this on other user's datasets too, including refractor, and with a sufficient star field the same problems arise, so it's not just my Newt.

As an extra bonus, depending on image scale, some of these AltStar profiles can become rather large (as in mother-of-all-starfish), exceeding the 25x25? max blue box of SVD.

[admin]

Alas, I must admit defeat in my attempts to use starfish for SVD sampling.

The theory seems sound, but in practice, the starfish are just too interconnected to get separation for sampling. It would take a lot of switching back to Mask, but I suppose I could use single-pixel-off to carve a ditch between merging profiles (requires some guesswork) and then erase the next-door star(s).

I have tried this on other user's datasets too, including refractor, and with a sufficient star field the same problems arise, so it's not just my Newt.

As an extra bonus, depending on image scale, some of these AltStar profiles can become rather large (as in mother-of-all-starfish), exceeding the 25x25? max blue box of SVD.

The solution is to shrink the sampled area, not increase it. You don't need to encompass isolated starfishes in the sampled area, you just need to find isolated stars with starfishes big and even enough. The sampled area (the squares) around the stars can be smaller than the starfishes themselves, no problem.

E.g. if this is your field of starfishes;

Selection_752.jpg (109.75 KiB) Viewed 4407 times

...then this is a reasonable way of sampling them;

Selection_753.jpg (110.65 KiB) Viewed 4407 times

BTW, I love it how you just coined starfishes as an official technical term.

Does that help?

[Mike in Rancho]

Well, technically it's starfishies. For greater irreverence.

But oh my goodness, yes that does help. I guess I have always been under the mistaken impression that the Law of Startools mandated the entirety of the sample's white outline to be within the blue box.

Now, my images seem to have far denser herds (schools?) of starfishies than your provided example, but I now get the picture and can begin experimenting with this in SVD.

[admin]

Well, technically it's starfishies. For greater irreverence.

Noted. Apologies.

But oh my goodness, yes that does help. I guess I have always been under the mistaken impression that the Law of Startools mandated the entirety of the sample's white outline to be within the blue box.

Now, my images seem to have far denser herds (schools?) of starfishies than your provided example, but I now get the picture and can begin experimenting with this in SVD.

Good to hear! FWIW this was one of your datasets, though I did make some changes in the development version when generating the starfishies, so it's possible mask generation is behaving slightly different...

Cheers!