Progress towards a high resolution deep colour image of M3

[timh]

Bortle 7 skies (partly astronomical twilight)
VX12 300 mm F4 Newtonian
Ioptron CEM70 mount
AS1294 MM camera. ZWO RGB filters. RGB captured with 4.63 uM pixels (0.81 arcsec/pixel). Luminance with unlocked 2.315 uM pixels (0.405 arcsec/ pixel)



Unsteady skies (PHD2 guiding no better than 1.3 to 2 arcsec so switched off) and breezes meant that imaging at home during the latter part of May was limited to a strategy of selecting from many short exposures. This is probably more feasible at D = 300 mm and F 4.0 than with smaller longer scopes.

About 100 x 10s gain 124 red, green and blue (ZWO colour filters) frames at -5C totalling 50min exposure – selected over several hours using SharpCap 4 live -stacking and FWHM + brightness filters - were integrated and combined into an RGB image at an image scale of 0.81 arcsec/ pixel using PixInsight. The FWHM values of the linear integrations of each channel were about 2.5 arcsec - use of Blur Xt (deconvolution) star shape correction and slight reduction improved these average values down to about 2.2. After stretching and processing these data produced a fairly sharp and colour vibrant image of the M3 globular cluster.

The star resolution was much improved – resolving M3 more or less to the core – when this initial image at 0.81 arcsec/ pixel was aligned and the luminance transferred from a luminance filter image integrating 1100 x 3s frames at a scale of 0.405 arcsec/ pixel. The linear 1100 x 3s integration was at an FWHM of around 1.65 arcsec (from about 1.9) after sharpening with BlurXt. However although adding the finer scale luminance data improved resolution it also had the effect of somewhat weakening and diluting the colour depth.

So - as always - room for improvement with more data.


(One thing to note of possible interest. Under such unsteady skies one of the biggest apparent problems was star-shape distortion - measurable as high Eccentricity values. It was notable not only that use of RC BlurXt star-shape correction was pretty effective at recovering improved star shapes but also that this process worked best when it was applied to integrations comprising the largest total numbers of frames notwithstanding some of them being poor. In particular pre-removing (and thus reducing total frame numbers by a half or more) by using PI subframe selector to omit the more blurred and eccentric frames actually led to a poorer rather than a better final result in most cases.)
Tim

[admin]

Hi Tim,

thanks for sharing both the images and the detailed description of how you achieved them.

Interesting that the deconvolution worked best with a larger number of total frames - my suspicion is that it will be related to the lower noise level if you stack more frames. Sharpening algorithms of all kinds tend to be limited in how much sharpening can be used by the noise level of the image, including both the unsharp mask and Weiner deconvolution in SharpCap.

cheers,

Robin

[turfpit]

Tim

A good image of a globular cluster right down to the core. Thanks for the write-up. A useful description of how to manage poor imaging conditions such as lack of astronomical dark, wind and unsteady seeing.

No astro dark here at 53°N for 8 weeks.

Dave

[timh]

Robin
[Interesting that the deconvolution worked best with a larger number of total frames - my suspicion is that it will be related to the lower noise level if you stack more frames. Sharpening algorithms of all kinds tend to be limited in how much sharpening can be used by the noise level of the image, including both the unsharp mask and Weiner deconvolution in SharpCap]

Thanks Robin,

I am sure that you are right about this - the more information the better the deconvolution works.

Overall seems to me that CMOS cameras and now routine use of deconvolution in processing is changing the old rules for how to do astroimaging

i.e.

It is no longer true that acquiring deepsky data at much below 1.0 arcsec/ pixel is to oversample because any subsequent application of BlurXt or other deconvolution requires higher sampling rate information to deliver more detail as per Nyquist (and in any case you can always recover SNR by binning in processing). Blur Xt brings only slight improvement to images sampled at 1 arcsec/ pixel -- because at that sampling it is hardly possible to expect FWHM values far below 2 and works far better at 0.5 arxsec/ pixel. But equally - as you say - deconvolution also requires decent SNR - which is a sort of catch 22 because it means imaging for that much longer (4x) at the higher sampling rate to get equivalent SNR.

My philosophy thus far - especially under unsteady imaging conditions - has been to accumulate lots of frames and then prune them by about 2/3 to select the best Eccentricity and FWHM (+ Star count) frames. Without any subsequent deconvolution this practice has worked to get sharper images. But - with routine use of Blur Xt in processing - it is definitely counterproductive and better to cull less ruthlessly and retain more information and SNR for deconvolution to work with. In case of any interest herewith some data...

Overall there were about 650 3s frames acquired. The lower left image shows a simple Sharpcap 600 x 3s autostack that comprised most of them - and associated FWHM (pixel) and Eccentricity values. To the right is the same image but after application of BlurXt star correction deconvolution .

At the top is a similar pair of images - also before and after deconvolution star correction- comprising just the 'best' 410 frames from the same set pre-selected using subframe selector prior to integration. Relative to the unselected autostack the FWHM of this integration is significantly improved and the Eccentricity slightly so.

However - looked at after deconvolution it can be seen that the preselection process ultimately gained me nothing at all. In fact I lost a lot because the best image of all --also with the highest SNR - is where I simply took the 16 bit autostack and applied BlurXt.

So in future will definitely be culling less and preserving SNR whilst continuing to 'oversample' luminosity

Tim

[timh]

Thanks Dave ,

Well at least I can console myself with the fact that down here in the deep south at 51.5 degrees the astro-twilight is only about 6-7 weeks. As I think you have seen I reworked the original image to take advantage of more of the data acquired (as per my note to Robin) and put it onto astrobin.

https://www.astrobin.com/7xe62n/

the additional data improved it a bit I think...

Tim