SharpCap Forums

VX12 Orion Optics (UK) Newtonian (f = 1200mm, F4.0), SW parracor type coma corrector, CEM70 Ioptron mount, Baader steeltrack focuser, Pegasus Cube2 focus controller, PHD2 guiding using an ASI 120 mm guide camera and 80 mm SW startravel refractor at f = 400 mm.

OSC image was 7 x 20s plus 100 x 5s exposures at gain 124 using an ASI 294MC camera - pixel size 4.63 uM

Mono image was 213 x 3s at gain 285 using an ASI 294MM camera - pixel size 2.315 uM

Bortle 6 moonless skies

Again the idea here was to take the OSC image and to try and sharpen it up by 1) substituting in the luminance from slightly sharper 3s mono frames and 2) doing the same but using a deconvolved 3s mono image.

The image on the left is the original OSC image, the central image is after addition of 3s image luminance and the one on the right is the same but after first deconvolving the mono image.

On this occasion the linear integrated mono 3s image was not especially sharp (FWHM ~ 2.25) and barely any better than the (also short frame) OSC image (FWHM ~ 2.4) . Not surprisingly, by itself, the addition of the mono luminacne resulted in only a correspondingly slight improvement. However deconvolution worked well to sharpen the image and made a big difference.

This type of bright high SNR and also detailed object is probably the ideal subject for deconvolution the success of which process is highly dependent upon and limited by noise. I attach a link to an interesting reference on deconvolution - and how it is used professionally alongside adaptive optics. Not pretending to understand all the math but sort of get it as an iterative fourier transform -- wavelet adjustment according to the (measured) image point spread function - and then fourier synthesis process ?? https://cfao.ucolick.org/aosummer/book/ ... _tyler.pdf

Anyway it still rather looks like magic to me - something for nothing -- or rather like seeing an out of focus image snap into focus.

I look forward to doing this target 'properly' sometime starting from some much better short frames ...

Tim

Nice job Tim!
I don't have a mono camera, so will this process also work capturing short exposures, process that as greyscale with some extra sharpening?

Menno

Hi Menno,

Yes indeed and I have used it on OSC images as well. One procedure (in Pixinsight) is as follows

1) Take the linear integrated OSC image to an early stage of processing - just as far as gradient subtraction and colour correction (optionally).

2) Change (or make sure) that the 'RGB workspace' of the image is linear i.e. sometimes the default does not have 'gamma' set to one so that just means making sure that it is set to one.

3) Then extract the CIE X Y and Z channels. The luminance channel is the Y channel and that is what you work on and deconvolve before recombining it back with the other two channels again to regenerate a colour image.

4) Next you need to establish a point spread function (PSF) that characterises the particular distortion prevalent in your image. It can be done by a 'parametric' method wherein you input guesses wrt typical distortions. PI offers a 'dynamic ' method which seems better where you pick 20-30 unsaturated stars from your image and then it measures a PSF value for you.

5) Then you need to think about masking. Build a transparency mask that only includes the high SNR areas that you want to deconvolve (low SNR areas just generate a mess otherwise). Also - to avoid ringing artifacts - you will probably also need another blocking type mask for stars outside of you area of interest.

6) Then - with the above tools as input - in PI you just run the deconvolve function. There are a few variables to play with (weight of deringing and limiting brightness) that you change in response to seeing various sorts of artifacts but it is quite straightfoward. For the he Eskimo nebula I used 100 iterations but improvements were obvious after the first 10 or so.

That will give you a sharper deconvolved Y channel which can then be recombined back into the colour image --and then processed as normal.

Generally I have been doing things slightly differently - in some cases deconvolving both the mono short frame image and the extracted Y channel and then recombining these two into a high dynamic range luminance which I then substitute back into the RGB image after stretching.

Tim

Thanks for the explanation Tim!
Normally I don't do this but in this case I'm going to put it all in Google Translate
A question though: you wrote "extract the CIE X Y and Z channels". What is CIE and what do you mean with X, Y and Z channels?

Menno

Hi Menno

Sorry I really didn't mean to make it mysterious (or no more mysterious than it is to me). The problem for me is that while I know how to do things within PixInsight I really don't know how those processes transfer to other software. Secondly it is all new to me too and so my understanding is shallow.

But the principles have to be the same whatever the software ...i.e. you need to establish a suitable PSF function and if you are going to sharpen the luminance only (which seems sensible because that is where all the detail is) then you have to be able to extract that out of the OSC camera image, restore it's sharpness, and then put it back in again.

There are a number of standard ways of mapping colour and luminance onto 3 orthogonal axise and it just so happens that the CIE XYZ system is one of them....CIE L,a, b and CIE L, c, h are two others for example. As I dimly understand it these systems interact with something called the 'RGB workspace' of an image which defines in a quantitative way how much of each of the 3 colour channels contributes to luminance number. For example you could do it evenly and have each contribute the same ...or -as I think is more common - have green contribute more to the luminance number than R and B. Then I am guessing that gamma would refer to how linear or non-linear the relationship between the calculated luminance and colour channel values are.

The CIE XYZ colour space is one of a number of colour spaces and happens to be the one that you use within PixInsight when you are doing deconvolution of OSC images. It maps color and luminance as in this link.

https://mymusing.co/cie-rgb-and-cie-xyz-color-space/

I hope that this helps ...not easy stuff

Tim

Thanks again Tim!
Makes a bit more sense now. And thanks for the link.

Menno