Hi Nikolay,
The result looking good, but bright stars burned out during deconvolution. It's normal or deconvolution settings is wrong?
It is completely normal. Deconvolution will tend to saturate all bright objects. Think of it in the following way: what is the "ideal" image of a star? quick answer: a point (well, an Airy disk in the real world, but we are speaking in purely abstract terms). Deconvolution will try to concentrate the whole flux of a star's image into something as close as possible to a point, provided your PSF is reasonably accurate. That's why the image of your star is smaller after deconvolution, and that's also why it is saturated: the same flux has been concentrated into less square pixels.
The dynamic range extension feature of Deconvolution can be used to palliate this problem to some extent. By increasing the high range extension parameter, Deconvolution will have more room (that is, a wider dynamic range) to accommodate brightened pixels. The price to pay is a loss of dynamics in the final image: the resulting dynamic range will be larger, which yields a darker image. This can of course be fixed with a nonlinear transformation.
The best way to prevent saturation problems, along with dynamic range extension, is to protect the brightest stars with a suitable star mask. As you only need protection for really bright stars (usually), star masks are relatively easy to build for deconvolution.
In the same way, a mask may be necessary to restrict deconvolution to high SNR regions, that is, to protect the background and transition regions between low and high SNR areas. This is typically a luminance mask. Our implementation of regularized deconvolution is extremely efficient so many times these background masks are not really necessary, or can be quite permissive.
You can combine a luminance mask and a star mask very easily with PixelMath (usually by just multiplying both masks).
No, no, no. In an ideal world, we we get image of Airy Disk.
Correct. The diffraction figure is the absolute physical limit that defines the theoretical spatial resolution achievable with a given aperture. Naturally, we never get something similar to the Airy disk as the PSF of a deep-sky image, since atmospheric turbulence and instrumental imperfections are always much larger. A "perfect" spatial telescope, as Hubble if it were "perfect", would be a nearly diffraction-limited instrument.
But now i totally don't understand how to choose "Shape"
The shape parameter defines the
kurtosis or
peakedness of the PSF. When shape is 2, we have a normal (Gaussian) distribution. When shape is less than 2, the PSF has a
leptokurtic (peaked) profile, and when shape is greater than 2 the PSF is
mesokurtic (flat).
Always think in "inverse terms" when applying deconvolution: Deconvolution will try to
undo the smearing caused by a previous convolution with the PSF. So if your PSF is leptokurtic (peaked), deconvolution will tend to cause less sharpening (or small-scale edge enhancement) than if you apply a mesokurtic (flat) PSF.
Obviously, this discussion assumes that you're finding your PSF by manual trial-error work. For deep-sky images, the PSF is usually fitted to a Gaussian (shape=2), as this is the "natural" probability distribution used to model atmospheric seeing phenomena.
Non-Gaussian PSFs are more useful to deconvolve high-resolution lunar and planetary images. For example, a peaked PSF can be used to approximate the profile of an Airy disk. The shape parameter allows you to control the effect caused by deconvolution on critical image structures. This is particularly important for lunar images, as I explained in
a specific tutorial.
