2023-03-15: APP 2.0.0-beta14 has been released !
IMPROVED FRAME LIST, sorting on column header click and you can move the columns now which will be preserved between restarts.
We are very close now to releasing APP 2.0.0 stable with a complete printable manual...
Astro Pixel Processor Windows 64-bit
Astro Pixel Processor macOS Intel 64-bit
Astro Pixel Processor macOS Apple M Silicon 64-bit
Astro Pixel Processor Linux DEB 64-bit
Astro Pixel Processor Linux RPM 64-bit
Better Qualitiy Weight through changes in Tab 3&4
Hello Mabula, Vincent and Wouter. Would it be possible to deactivate the Star Target in Tab 3 so that all stars in the image are detected. And to limit a Star Target in Tab 4 for the stars used for registration.
The background is that if all stars in the image are detected, the quality weighting works better. Because then images with clouds or high light pollution are better downgraded in the quality ranking. Currently, the Star Target (default 500 stars) in Tab 3 for widefield images with thousands of stars makes the Star Density the same for all frames.
So frames with (moonlight, clouds, light pollution) get a high value (small stars and high SNR) and the resulting fluctuations in the Star Density remain uncovered for APP. Because they are above the star target.
If I turn off the Star Target in Tab 3, the registration becomes very slow or does not work because there are too many stars to process.
For small fields of view with only a few hundred stars in the image, the quality weight works very well. But with widefields, due to the above-mentioned restrictions, it does not.
Adam Block had once presented in a video such a formula, where SNR and star number were related. So if for example by clouds the SNR went up and the number of stars down because covered by clouds the images were penalized in the weighting. But for this all stars must be found in the image.
I think such a change in APP should not be difficult to implement.
With kind regards, Henry.
No one has an idea about this or opinion?
Sorry for the late respons, human malware and holidays were delaying things overhere. I'll ask Mabula about this one.
Hello Mabula, I took some screenshots to make it clearer.
In picture one you can see the best qualitative image after the star analysis with standard settings.
In the next image, the reference image automatically selected by APP. It shows a gradient, probably caused by thin clouds or light pollution.
And now you can see the analysis chart sorted by recording time for all sessions. The Star Density becomes useless as a weighting criterion, because the same number of stars is detected in almost all images and only the FWHM value is left for weighting. But this can be easily deceived by clouds, because then stars appear smaller.
If I now deactivate the star target in tab 3 and all or many more stars are detected. The results are much better, as in the next image. With the choice of the refernce frame.
And the best quality image according to the star analysis.
And still the quality history of the sessions sorted by recording time. Reflects the actual quality weight much better when the images then looks at.
The quality curve is more like the curve of the stars found.
You would now only need to disable the Star Target in Tab 3 and only add a Threshold switch in Tab 4. Where you define how many stars are used for registration, but for weighting all found stars are used. In order not to slow down the registration too strongly with many stars in the picture. I think this should be feasible. And in the quality score Star Density and SNR can play a bigger role and FWHM only a small part.
This idea came to me in the above video of Adam Block.
I still add the link.
Even if the whole thing is demonstrated there in Pixinsight, it should be applicable in APP.
With best regards Henry.
Thanks for the added details, very interesting and I forwarded it to Mabula.
I assume that you have completed some integrations of the same target using quality based weighting and compared the results of using default (500) and much higher numbers of stars during star analysis?
If so, how well do the results compare?
Like you I long ago noticed that running star analysis with a high star target resulted in much higher quality weights being assigned to the subs compared with running with a low target. However when I compared the results of the comparable integration runs I was unable visually or using the noise , SNR etc statistics assigned to the integrations by APP to actually perceive any convincing improvement.
If you have done such a comparison, how well did the results compare and on what measure did you decide there was actual improvement?
After a few trial comparisons, as I could find no noticeable improvements, I concluded that although using higher star detections resulted in generally higher quality scores there was not any significantly better discrimination between higher and lower quality images at least in terms of the eventual results. I accept from your initial post that APP with a higher star target seems in your example to have selected a better quality reference frame, but unless there is a general reduction in the Registration RMS numbers is this an overall improvement if the scaled quality weights used during Integration are almost unchanged?
The images I shoot are typically 1.1 x 1.4 arc degree so not wide angle. Maybe the effects of higher star targets are more noticeable on wide angle or some other feature of your approach versus mine.
Anyway I repeated my earlier trials using some recently acquired 2x1 mosaic images of the Soul nebula. Visually I cannot see any obvious improvement between the two sets of results.
Looking at APP's analysis of the integrations I see that the results of the high star target seems to consistently have lower noise levels (which seems good) but the SNR figures are virtually unchanged so not sure if there is actually a measurable improvement overall. My results are below. Maybe someone with a deeper understanding of APPs numbers can see something in my results to convince me that integrations using the higher target star numbers are materially better? Or suggest a method that would convincingly show an improvement or otherwise? Although I set the higher target at 10000, APP was only actually detecting c3500 stars at best.
Star Target Filter Frames BG-1 Scale-1 Noise-1 SNR-1
500 Blue 26 1.5330E-02 3.5660E-04 1.5151E-04 4.1645E+00
10000 Blue 26 3.4669E-02 2.2715E-04 9.2468E-05 4.1215E+00
500 Green 37 1.5492E-02 4.7886E-04 1.3535E-04 5.4269E+00
10000 Green 37 3.4772E-02 2.8748E-04 7.7582E-05 5.4941E+00
500 Ha 59 1.5439E-02 4.9600E-04 1.2790E-04 5.2099E+00
10000 Ha 59 3.4755E-02 3.0654E-04 7.3998E-05 5.2575E+00
500 Red 38 1.5420E-02 5.0393E-04 1.4094E-04 6.0648E+00
10000 Red 38 3.4739E-02 2.9800E-04 8.1934E-05 6.1578E+00
Star Target Filter Frames BG-1 Scale-1 Noise-1 SNR-1
500 Blue 30 1.2903E-02 3.4672E-04 1.2507E-04 4.9848E+00
10000 Blue 30 3.3416E-02 2.0526E-04 7.3840E-05 5.0417E+00
500 Green 38 1.3001E-02 4.3208E-04 1.2527E-04 5.6628E+00
10000 Green 38 3.3456E-02 2.4964E-04 6.9927E-05 5.6567E+00
500 Ha 73 1.2871E-02 4.5004E-04 1.0376E-04 5.9722E+00
10000 Ha 73 3.3386E-02 2.4842E-04 5.5405E-05 6.1612E+00
500 Red 30 1.2985E-02 3.8418E-04 1.4127E-04 5.3988E+00
10000 Red 30 3.3458E-02 2.1594E-04 7.8287E-05 5.5948E+00
In conclusion, I accept that my one set of results is hardly statistically significant but I am doubtful your proposed change will see dramatic improvement in the resulting integrations. Would love to be proved wrong. And if the proposed change is straightforward I guess even a modest improvement is a step in the right direction.