Відео

Thanks! The pixel size is actually around 4.63 microns. Bin 1 is about 2.3. Bin 2 is 4.6. For some reason Siril (and other program) think the pixel size is 9.2, which is wrong. Not sure why. Luminance is 120 seconds taken with an F/6 scope (the equipment is in the description of the video I think). I probably should have shortened the exposures a bit. But, live and learn!

I know! I've gotten so used to it.

It is easy! Quick question: was chat gpt accurate?

Entry submitted. I remember now why I was so excited to get Blur Xterminator - and why I never again (until now) touched the regular deconvolution tools in PI

Too be fair, 90% of us consider it artwork to a certain extent. Some people freehand paint on their images, some use wild colors, some use AI. It's really dealers choice, we all have our own personal tastes, and that's okay :)

Thanks! Just checking, have you seen the follow up video? ua-cam.com/video/sEpPQXq5TIY/v-deo.html

@@deepskydetail Oh there is a follow up video, Watching it now ! Thanks :)

Can't wait!

@ if it’s ok with you - Can I also make a video about the process?

Of course!

I think you can do it! Basically, most tools in Pixinsight (except add ons like BlurX and cosmic clarity) should work. Same for Siril and Gimp. Most everything in photoshop too I would think.

@deepskydetail does that include the denoise and dehaze functions in the Camera Raw filter of Photoshop? Have just started using Siril.

I'm not sure about dehaze. I think with camera raw there are two denoising tools. The older one is fine. The newer one uses AI.

Yeah, deconvolution can be slow. I've found that Siril's deconvolution tool is not to bad with a food pc. But it still can take a while.

@@deepskydetail I bought pixinsight and after that never touch sirril again. Not only that those days were no joke traumatic for me. I burned all my patience those early days when I was trying to overachive with my puny gear.

Thanks! Looking forward to seeing the submission!

The best school... ? :)

@@deepskydetail 😁 Not according to some of the facebook groups I'm in

Thanks! I hope it turns out to be a fun project :) If you need any pixelmath tutorials, Stefan Astro has a good one :) I look forward to seeing the submission!

Thanks :)

Thanks for the comment! You may not need to go more than 15 seconds depending on the speed of your scope and how much light pollution you have. The main thing is to get past the read noise :)

Glad you've found it and glad you found the video useful :)

Thank you!

Thanks! And thanks for the sub! That's a good question. I'm not 100% sure what the best answer is. But I think there are two decent answers. I would go with option 1 if you're imaging from light polluted areas, and option 2 if you're in relatively dark skies. [1] just use shorter exposures (e.g., 120 seconds or 60 seconds). If the core is getting blown out, shorter exposures will (probably) still be above the read noise for the fainter areas (M31 is pretty bright). Or... [2] as you stated, mix short exposures with longer ones, then combine them later on in photoshop/Gimp. You could probably even stack the short and long subs together using Siril/Sirilic, Pixinsight or DSS, making sure to separate the sessions (short vs. long subs). But, experimentation might be needed to get it right.

Assuming tracking is smooth an unaffected by winds etc

Thanks! That's a great question. I'd have to think about how to test that. Any suggestions are welcome!

You're not the only one! 😅

Yes! And they take up a lot more space on your hard drive!

Thank you 🤗

Thanks for the comment. I'm not exactly sure I understand what you're trying to say. The point I was trying to make is that shot noise from light pollution is random, which destroys SNR. It destroys faint signal's SNR faster than bright signal's. You can't just subtract light pollution from an image to restore the SNR because of the randomness. Longer subs won't help.

@ I see what you mean, but the signal is just as random as the light pollution. Both are noisy. The noise is due to the random nature of photons hitting the photodetector. This noise is square root of the number of photons so if the number of photons is large, the signal to noise for both the light pollution and the signal increases. At long exposures, the signal to noise ratio can be large enough such that light pollution (assuming it is uniform across the field of view or can be somehow approximated) can be simply subtracted out.

Unfortunately, I don't think that's the case. The DSO signal and LP signal are not equally noisy. In fact, the light pollution signal is generally much noisier. Exposing longer increases both the LP noise and faint signal noise. Its randomness makes it so you cannot approximate it and subtract it out (i.e., the LP noise and DSO noise are not correlated).

@@deepskydetail both are weak light signals, and consequently both have intrinsic shot noise. Both light signals are registered by the photodetector. How would photodetector know which is which? When the light pollution is much stronger than light from DSO, its shot noise is also stronger so it may overwhelm the DSO signal. Perhaps this is what you have been observing. However, fundamentally, both light signals follow the same rules. Theoretically, if you could integrate forever, you would be able to pull out even a very weak DSO signal. I would suggest that you read this: en.wikipedia.org/wiki/Shot_noise

I'm still not sure what you are trying to say. Are you saying, "stacking longer sub frames is better than stacking shorter sub frames?" That's what your original comments sounded like to me. Or is it "it's only total integration time that matters?" What does "can only be subtracted based on our subjective judgement" mean? Maybe we're saying the same thing, but in different ways. The point of the video is that in light pollution, once you get past the read noise, the length of the sub frames does not matter. At that point it's only total integration time that matters. But still, light pollution destroys faint SNR because, as you correctly stated, the camera cannot tell the difference between the DSO and light pollution. For each photon that hits the sensor from the DSO, a random number of photons hits the sensor from light pollution. That destroys structure. The SNR never goes to zero, so you can, in theory, keep stacking to average out the randomness. But, the SNR is much less than in dark skies. The signal does get overwhelmed. That's why in a Bortle 9 zone you need orders of magnitude more integration time than in a Bortle 1.

Thanks! Yes, I don't think it will be very representative. Still trying to figure out how to measure blotchiness better! Any suggestions are welcome!

Thank you! I appreciate the support!! :)

Thanks, Naz!! Glad to know the data match your experience :)

Thank you! Totally agree: dual rig setups seem to have a lot of benefits :)

Agree. There are so many variables to think about when choosing a good exposure time.

The precision of the 32 bits, which is a floating point so actually only a 24-bit mantissa, comes from the stacking. This averaging of all of these discreet values creates new intermediate values. You actually gain levels by a factor of log2(N) where N is the number of lights (subs). It’s easy to visualize because with two subs you can get a value of 20 and 21 and this will average to 20.5 which is an intermediate value so you double the number of intermediate values and now you have halves. With four subs you have quarters you can have the results of 20.25, 20.5, 20.75. So with 256 lights log base two of 256 is eight you gain eight bits of precision because you can now create 256 imtermediate levels. This is why 32-bit precision is needed during stacking, otherwise these fractions values generated by the averaging get truncated.

@@stevenmiller5452 yes I thought it could be 32-bit float - in PixInsight that would be the norm, but I’m not familiar with SiriL. Also the examples given were all unsigned integers, hence my question. But 32-bit unsigned integers would be IMO too limited to represent the result of stacking.

@@soaringfranzIt’s academic at this point but you could, in theory, use 32 bit integers, the stacker could use a summing method, instead of averaging, and then simply rescale to a 32-bit float (divide the final result by the number of sibs) afterward to save in the proper format. The nice thing about floats is it’s easier to do weighted stacking by applying a non-integer weighing factor for instance. Lot’s of math is just easier to program with floats.

@stevenmiller5452 agreed, but the fact remains that the example in the video performs an average in unsigned integer space, or all the discussion about rounding wouldn’t have any meaning

@@soaringfranz Yup, that was a problem. Good thing he caught it.