Curious about deconvolution

[N_DD]

Hi,

(not sure this is the right forum to post my question...)
I was wondering if there is an optimal setting for the Radius parameter (Decon module) linked to the image scale (arcsec/pixel) and seeing for deconvolving an image. If I am not mistaken, the Radius should somehow control the size of the point spread function (PSF) used by the module, and then one would like to have it tuned such that its size matches the seeing. Alternatively, it could be tempting to run some blind deconvolution on a smaller area of the image just to estimate the "true" PSF, and then use it for deconvolving the whole image...

Thanks!

Nico

[admin]

Hi Nico,

You're right in that Decon in StarTools is primarily geared towards countering seeing-related blurring, given that its point spread function approximates a Gaussian kernel. The Radius indeed controls the radius of the presumed Gaussian blur that needs reversing.

Hope this helps,

[N_DD]

Hi Ivo,

thanks: so I should start with a Radius which corresponds roughly to 1.5"-2" to be in the ballpark of seeing induced blur, isn't it? In my case, imaging at 0.92 arcsec/pix, this would mean 1.6 to 2.2 pixels... if I bin the image, I would end up with a Radius which is less than one pixel! Does deconvolution still make sense?! Am I doing anything wrong?

Any plan to implement blind deconvolution, or position-dependent deconvolution in Star Tools?

Thanks,

Nico

[admin]

Hi Ivo,

thanks: so I should start with a Radius which corresponds roughly to 1.5"-2" to be in the ballpark of seeing induced blur, isn't it?

Unless you're in the Atacama desert, chances are that 3-4.5 is more realistic - on a good night But that's the cool thing about the preview area - it'll allow you to quickly tweak the kernel until you've found the sweet spot.

In my case, imaging at 0.92 arcsec/pix, this would mean 1.6 to 2.2 pixels... if I bin the image, I would end up with a Radius which is less than one pixel! Does deconvolution still make sense?! Am I doing anything wrong?

Your are absolutely correct - binning is actually another way of making better use of your signal. In fact, binning a little bit may make the difference between data that is unusable for decon (due to noise at higher resolution) and usable (due to improved signal at lower resolution). You are also correct that data that is no longer oversampled (due to binning or due to a low-res camera or due fast optics), is no longer a candidate for deconvolution - if the aim is to use deconvolution for the reversal of seeing related issues.

The atmosphere is not the only contributor to the PSF - your lens and/or scope will necessarily diffract the light as well and it's always possible something else in your optical train or location is causing the PSF to be bigger.

It has to be said that deconvolution - as an algorithm/tool - can be used to correct for some other issues as well (not just seeing) such as tracking error or bad collimation, but that typically requires a kernel that is no longer (close to) Gaussian. The thing with those sorts of issues though is that they are much harder to correct due to the presence of singularities in the data (e.g. over exposed star cores). Gaussian kernels are nice and symmetric which makes it much easier to control for any ringing that such singularities cause.

Any plan to implement blind deconvolution, or position-dependent deconvolution in Star Tools?

Blind probably not (as its usefulness is limited in AP), however I have experimented with position-dependent and arbitrary Point Spread Function deconvolution in the past. For the time being, the conclusion is that deep space astrophotography makes it extremely hard to benefit from the latter two types of deconvolution due to aforementioned problems with overexposed star cores. Seeing-related blurring reversal is by far the most useful application of deconvolution and that's why the bulk of the R&D has gone into this area.

Hope this helps!