Why does SVDecon affect non-masked areas?

[BrendanC]

Hi,

I select the SVDecon Module, I auto-generate mask, it selects the stars... and then, the space between the stars seems to change too.

Why is this? I'd have thought masking just the stars should process just the stars?

Thanks
Brendan

[admin]

Masks can be used in many different ways. They are simply a means to tell a module or algorithm do/don't process this pixel or do/don't measure this pixel or process this pixel differently.

They are used to distinguish pixels (or areas of pixels) for some purpose.

For example;

• In the Sharp module, they are used to signify stars so that they can be processed differently (avoiding bloat commonly seen in less sophisticated wavelet sharpeners).
• In the SVDecon module, they are used as apodization masks to provide the module an outline of a PSF ("star") sample that is as "pure" as possible.
• In the Layer module, they are used to switch between background and foreground.
• In the Wipe module, they are used to exclude areas from being sampled.

Etc.

User-specified masks in StarTools are binary; on/off. Luminance masks, as found in other applications, also include how much a pixel should be considered/measured/removed. Luminance masks are considered hacks/sub-optimal in StarTools for the vast majority of use cases, so you will not find them in StarTools, except for the Layer module.

I hope this helps!

[BrendanC]

I think it helps, I'm not sure!

So, specifically, in SVDecon, are you saying that when stars are masked, these are just being used as samples for the algorithm, which then applies processing to the entire image? Because that would account for why the entire image is affected, even the dark space between stars, even though only the stars were masked.

It's not really a problem, it's just something I've noticed and wondered why it happens.

[Mike in Rancho]

It took me a little while myself in the transition from 1.7 to 1.8. As Ivo indicated they are for star sampling in SVD, but the star masks also help the deringing parameters do their thing. So sometimes when I have bad stars that the automask doesn't properly pick up, I'll go fix up the mask manually (even though it's not a star I would use for sampling).

In 1.7 yes there was more of a classic operate/don't operate mask regime, with an inversion if I remember right, and maybe because old deconvolution was more prone to artifacting certain stars. But now we're global!

[BrendanC]

Thanks again Mike.

In which case, my preference is the old regime!

I've spotted that after SVD, the entire image is 'lifted' somewhat. I don't like that my nice, dark space suddenly seems to have had some sort of gain applied to it. I do however really like the way it firms up my stars. So if there's any way to have it firm up the stars, and doing its magic also on galaxy/nebula regions, while leaving everything else untouched, that would be ideal.

[Mike in Rancho]

Thanks again Mike.

In which case, my preference is the old regime!

I've spotted that after SVD, the entire image is 'lifted' somewhat. I don't like that my nice, dark space suddenly seems to have had some sort of gain applied to it. I do however really like the way it firms up my stars. So if there's any way to have it firm up the stars, and doing its magic also on galaxy/nebula regions, while leaving everything else untouched, that would be ideal.

Yeah I've seen that happen too, though not just SVD can do that. HDR can, for the highlights. Sharpen, which I rarely use anymore, can do it. Even 1.7 deconvolution. For me it tends to happen more with noisy, weak datasets. Of course I specialize in those! If you take things through the end using SS and then denoise, tracking actually will catch a good bit of it. But when really bad I sometimes realize I have to just back down the SVD iterations a few notches.

[hixx]

Hi Brendan, Mike,
I think the whole point of SVD is to reverse blurring caused by Atmosphere and/or optical train. As the whole image is affected by such, consequently Dekon must/should operate globally, while deriving the local variations in the required "anti-blur"-model happens regionally/locally using the star sampling method. As Mike explained, the Apod mask is used for both:
1) Identifying sample candidates and for
2) Identifying potential Singularities (for Deringing
In 1.7 You used an inverse star mask to exclude singularities from processing (which is covered by "positive" Apod mask, bullet 2, (which might be confusing, as the mask indicates the "do-not-process" areas). I am not sure if there would be any point in using a mask to Dekon just parts of an image other than artificial "enhancement" of an area (which could probably be covered using the Layer module).

Clear Skies,
Jochen

[Mike in Rancho]

Thanks Jochen, I agree. I think it's really more a matter of getting used to 1.8 from 1.7. That's quite a while now for me, so I'm fully pleased with SVD.

But if you really got used to 1.7, well that mask usage was perhaps both good and bad. One could manipulate the effects of deconvolution perhaps more than would be truly warranted from a documentary standpoint, yet being creative with the masks could also get you out of some trouble - say with globs.