Software Binning and WBPP

scotty38

Well-known member
There was another thread broadly about this where I asked the question but I think it got lost so thought I'd start a new thread.

If I wanted to take advantage of binning but retain the convenience of the WBPP workflow how would I introduce. or rather what would be the best way to introduce, the Integer resample process?

I did have a look at the options in WBPP but I could not see how it could be pulled in as it stands but maybe I am missing something obvious.....
 
Hi @scotty38,

binning is already handled by WBPP. You don't need to do anything special, just add your 2x (I guess) binned files and they will be grouped together in both calibration and post-calibration groups. WBPP never groups files with different binning together but after registration, the scale of the images will be the same as the scale of the reference frame, so you'll be able to compose 1x with 2x binned images of the reference frame for them is the same.

If, differently, you want to keep the scale separated you have to use a dedicated keyword (maybe BINNING itself) and register "per keyword", this will use a separate reference frame for each binning keeping the size of the master lights separated.

So you have few options depending on the workflow you would like to implement :)
 
ok thanks Robyx but my question is based on the images coming out of my camera in let's call it their "normal" state. I have a QHY294M and I can choose to image with 4.63um pixels or 2.15um pixels. If I choose the latter I may want to bin later, 2x2, 3x3 or whatever in software so at what point do I do that?
Do I do it before the files touch WBPP?

Oh just to add this is a new camera, I've not used it yet hence the question and uncertainty :)
 
ok thanks Robyx but my question is based on the images coming out of my camera in let's call it their "normal" state. I have a QHY294M and I can choose to image with 4.63um pixels or 2.15um pixels. If I choose the latter I may want to bin later, 2x2, 3x3 or whatever in software so at what point do I do that?
Do I do it before the files touch WBPP?

Oh just to add this is a new camera, I've not used it yet hence the question and uncertainty :)
ah ok, I misunderstood the question!

I would apply the software binning to the calibrated images. I would acquire images at binning 1, use WBPP to calibrate them, downscale the calibrated images and add these to a new WBPP session to continue with the remaining steps.
 
Exactly what "advantage" are you looking for?

Maybe I should not have used the word advantage and maybe I should have asked another question instead so here goes.....

Let's say I have an Esprit 120, focal length 840mm and let's say I leave my 294m in 2.15um pixel mode and take a load of images.

On a quite good night my AZ-EQ6 will guide at about 0.7-0.8" rms. According to the figures the Esprit and 2.15um pixels is at 0.57"/pixel so I would be over sampled and would I not want to bin 2x2 at that point as everything I read suggests that I should but I will readily accept that I don't fully grasp why.
 
My standard warning first: I am not an expert.
I think the key factor is whether or not your pixels are over-sampling the image. If they are not (i.e. your PSF <~1.5 pix), then binning will dilute your signal data and probably degrade the image. You say you get ~0.8" guiding accuracy; I usually get a bit better than this, but since the seeing in the UK is rarely better than 1.5" this usually doesn't matter. At 0.57"/pix you are probably over-sampling by at least a factor of 2, so 2x2 binning will probably give you some SNR improvement without significant loss of real resolution. However, I would recommend that you leave this until after registration and integration.
 
Your calculation of the pixel scale (0.57 "/px) is correct.

However, in my view you should not compare pixel scale with guiding accuracy (0.7 - 0.8" rms), it must be compared with the average FWHM of the stars in a frame of the imaging camera.

Bernd
 
it must be compared with the average FWHM of the stars in a frame of the imaging camera.
I agree. When I said "PSF" I really meant the FWHM of the PSF (which will combine ideal optical resolving limit, tracking errors, seeing, wind vibration etc.)
 
ok thanks both but I am not sure if my question is answered sorry :)
Given this possibly hypothetical situation what do I do, do I just process away in WBPP as normal or do I do something different? Do I do some intermediate stage of integer resampling because of this very situation I am in? Fred mentioned to do it after integration but I think Robyx suggested a different stage to do it so I am still confused and definitely none the wiser.

Thousand apologies......
 
Even with an oversampled image, alignment is likely to be more accurate if you keep the higher resolution. It also means that the interpolation required during registration happens at the higher resolution. Once you are aligned, there is probably very little mathematical difference between binning before or after integration - provided that you really are at least x2 oversampled. My instinct is to hang on to the higher resolution as long as possible. The only advantage of binning (other than saving disk space) is a small improvement in SNR (if the residual noise were zero mean Gaussian you would theoretically gain a factor of 2 in SNR with 2x2 binning, but my experience is that you actually achieve much less than this; it may depend on how carefully you have calibrated out any systematic noise, such as dark current).
 
ok thanks, I really did think this situation was a bigger deal so in general terms what does everyone with long scopes and small pixels do?
 
I hardly ever bin my images. Partly, this is because, having spent good money on a 20MPix camera I hate to give up my pixels, but also because (perhaps as a result of having spent a lifetime in digital signal processing), I'm not quite as obsessed as some with the last decimal point of SNR. I'm willing to leave a little bit of noise for my well-designed (half a billion years of evolution) visual image processing system to work on. And, joking aside, I honestly think that pictures with a little bit of residual noise actually look better (what an awful heresy)!
 
Thanks Fred, I'll no doubt have a play and see what happens once I'm back up and running.
Hi. I have been considering exactly this issue (my camera and reflector gives me 0.33"/px which is oversampled by some distance). So I am curious as to which method you tried out and (more importantly) whether it improved matters?
 
Hi. I have been considering exactly this issue (my camera and reflector gives me 0.33"/px which is oversampled by some distance). So I am curious as to which method you tried out and (more importantly) whether it improved matters?
I nearly always run my camera hardware binned 2x2. That changes my sampling from 0.46" to 0.93", which is still oversampled for my usual seeing. I've experimented both ways, and I don't find that I detect any difference in what stars or fine structure looks like. So I stick with binning, which provides slightly lower noise in situations where I'm using a short exposure time, creates smaller file sizes and faster processing, and most importantly in my setup, significantly reduces my download time and therefore increases the number of exposures I can make in a given period. The binning also speeds up my autofocus runs, although I could run those binned while not binning my other images.

I think the takeaway is that every setup and situation is different, and the only real answer is going to come from experimentation.
 
The only advantage of binning (other than saving disk space) is a small improvement in SNR
Now that this thread has been revived, I should clarify this statement in post #13, in case it is misunderstood (I have found that folks sometimes get this wrong).
If you add together a number N of samples which include noise (which we will assume is approximately gaussian), the noise on the summed (integrated) sample is reduce by a factor of √N. So 2x2 binning (four samples) reduces noise by a factor of 2, and 3x3 binning (9 samples) reduces noise by a factor of 3.
... so if I apply this to 100 3x3 binned images, don't I get a 300 times reduction in noise? No. the 3x3 binning gave you a factor of 3; the 100 (=10x10) integration gave you a factor of 10; the whole process gave you a factor of 30.
But the key point I want to note is that you get exactly the same reduction if you do the 3x3 binning after the 100 frame integration. You can "cash in" your binning noise reduction at any point in the linear processing workflow. It is sometimes suggested that registration (with interpolation) invalidates this, but registration just redistributes the original errors (though it may produce additional rounding / quantisation errors).

Bottom line: if you are undersampled, keep all your precision and decide as late as you can in the linear processing workflow whether or not you want to sacrifice detail for a small noise reduction (the answer should usually be "no"). If you are oversampled, keep all your precison at least until you have completed any PSF-based processing.
 
Last edited:
Thanks again. I am now using Blur XT and it makes a nice job of tightening up flabby stars. So it may well be that I don't need any downsampling at all.

As you say, experimentation is key.
 
Back
Top