SubframeSelector

[Corries]

Can anyone please help?  I have been using SubframeSelector to choose what subs to use in the stacking process.  However, with the Ordinate, Median, I am unsure what this is measuring and what is desirable, the high number subs or the low number subs.  With respect to the screenshot attachment, should I keep or discard the first eighteen subs?

Many thanks, Terry.

[John_Gill]

Hi,

Well it depends ... Median refers to the "Electrons or Data Numbers in a subframe, so I guess the higher numbers are better.  I first use "Blink" to remove oval star images and images with bright satellite streaks as this can affect SNDR values.  I then use SubFrameSelector and use SNRWeight and Eccentricity to evaluate an image.  If I have a number of images with similar SNRWeight then the best Eccentricity wins.  So re-sort your batch of images according to SNR and that should give you a better idea of what to keep.  The other thing is, do you really want to dump 18 images (50%) .... hope this helps.

Look up
John

[dld]

In widefield images, higher median usually mean higher light pollution or bright passing clouds. A passing airplane may spike up the median values too. In my situation, the median falls as my targets move towards the zenith, where LP is less prominent so I prefer frames with lower median. In your case it might be something else. Are your images from 2 different nights? Did you change your exposure settings? Visually, what is the difference between frame 18 and 19?

[Corries]

Thanks for the replys.  These are not widefield images but, yes, image 18 and 19 were on different nights but visually I cannot see any difference between them.  The exposures were the same for all of them.

[bulrichl]

Hi Terry,

given that you compare images of the same region of the sky I consider it quite unusual that there is - consistently! - a factor of 5 between the medians of two nights. Are you sure to compare images with the same pre-processing (calibration, optionally debayering)? And are the images really taken by the same acquisition software and same settings?

Bernd

[Corries]

Hi Bernd,

Thanks for the reply.  Same acquisition software but the same settings?  I cannot say as the time between the first 18 subs and the rest was considerable.  However, thanks everyone for your help.  I think you have answered my main question with respect to the Mean Ordinate.  The higher the Data Numbers the better. 

Thanks again, Terry.

[ngc1535]

Hi,

Well it depends ... Median refers to the "Electrons or Data Numbers in a subframe, so I guess the higher numbers are better.  I first use "Blink" to remove oval star images and images with bright satellite streaks as this can affect SNDR values.  I then use SubFrameSelector and use SNRWeight and Eccentricity to evaluate an image.  If I have a number of images with similar SNRWeight then the best Eccentricity wins.  So re-sort your batch of images according to SNR and that should give you a better idea of what to keep.  The other thing is, do you really want to dump 18 images (50%) .... hope this helps.

Look up
John

Hi John,

This is interesting. I understand that frames that contain bright satellite trails (for example) make for a "noisy" image- but are you permanently not using those images? Proper rejection (with enough dithered frames) will completely take care satellite trails...especially *bright* ones. I guess I have this fear you are throwing away otherwise good images.
-adam

[bulrichl]

I think you have answered my main question with respect to the Mean Ordinate.  The higher the Data Numbers the better.

I don't agree. My experience in this respect is like that what dld wrote:

In widefield images, higher median usually mean higher light pollution or bright passing clouds. A passing airplane may spike up the median values too. In my situation, the median falls as my targets move towards the zenith, where LP is less prominent so I prefer frames with lower median.

However, a passing airplane would not alter the median of the whole image seriously - the fraction of pixels is too low. Besides only few images would have been affected by airplanes and not all images of one night.

Like John I would take a look at the SNRWeight of images 1 - 18 (group 1) vs. images 19 - 52 (group 2). In my opinion SNRWeight is the criterion that is the much more important than the Median. IF the higher Median of group 2 images was caused by light pollution (e. g. moonlight) or passing clouds, the SNRWeight of these images would be considerably lower than the SNRWeight of group 1 images. You can easily check that.

However, I can't imagine that this is the cause. There is almost no variation of the Median within one group of images. So I think this indicates some other reason (e.g. different pre-processing or different conditions at image acquisition). If you want I'll take a look at one image of each group (XISF files).

Bernd

[RickS]

Moonlight or light pollution reflected by passing clouds will typically *increase* SNRWeight.  I always check and discard obviously bad subs using Blink then use SubframeSelector for a second pass evaluation.

Cheers,
Rick.

[bulrichl]

Hi Rick,

I am referring to that what dld wrote:

In my situation, the median falls as my targets move towards the zenith, where LP is less prominent so I prefer frames with lower median.

This is also what I am seeing.

I just took a look at my last imaging sequence, whose Subframe Selector output was saved as CSV file: with diminishing Median, the SNRWeight increases. So these values go in opposite directions. This is logical as well: the decreasing background level has lower shot noise, making for a higher SNR overall. Really don't know why it should be the other way.

Bernd

[dld]

Hello, Bernd, friends,

I live near an airport and occasionally, airplane tracks can cover a considerable amount of my frames, increasing both the Median and SNRWeight values. Sometimes airplane spotlights may lighten up an entire frame without leaving any other tracks .

[bulrichl]

OK, you're right, one should not generalize... I live on La Palma (Spain, Canary Islands) and this region is a no-fly zone at night. So fortunately this issue is not my concern. 

Bernd

[RickS]

Hi Bernd,

I just took a look at my last imaging sequence, whose Subframe Selector output was saved as CSV file: with diminishing Median, the SNRWeight increases. So these values go in opposite directions. This is logical as well: the decreasing background level has lower shot noise, making for a higher SNR overall. Really don't know why it should be the other way.

Not sure what is happening in your case, but more signal gives better SNR.  It doesn't matter whether a large component of that signal is unwanted light pollution.  Shot noise increases as the square root of signal, so the more photons we measure the better the SNR.  Unfortunately, our cameras can't distinguish between good photons and bad ones 

Cheers,
Rick.

[bulrichl]

Hi Rick,

Not sure what is happening in your case, but more signal gives better SNR.  It doesn't matter whether a large component of that signal is unwanted light pollution.  Shot noise increases as the square root of signal, so the more photons we measure the better the SNR.  Unfortunately, our cameras can't distinguish between good photons and bad ones 

Background is background and not the desired signal. Subframe Selector estimates both background and signal and provides the quotient, SNRWeight.

Take a look at this thread:
https://pixinsight.com/forum/index.php?topic=6746.msg45614#msg45614
Mike Schuster, the author of Subframe Selector, has explained in clear words what happens with SNR when light pollution or sky glow increase:

PI's noise evaluation tends to measure noise in a frame's background. This goes up with increased dark current (e.g higher sensor temp), increased light pollution, increased sky glow, etc. Hence SNR tends to decrease in these cases.

The signal part of SNR is a measure of frame scale or contrast. This goes down with poorer transparency, thin high clouds, etc. Again SNR tends to decrease.

Both noise and signal measures are estimates, so there is room for error. For example if the frames have differing gradients the estimates can be less accurate.

This is my understanding as well, and it is also what I am seeing in my data. I don't understand what you are saying.

Bernd

[RickS]

Hi Bernd,

I don't understand what you are saying.

I'm saying that bad subs can have high signal and hence good SNR and that it is unwise to rely on SubframeSelector as your only tool to winnow out bad subs.  You need to inspect them visually as well.  I have frequently had runs where cloud affected subs had SNRWeight better than many of the good subs.

Cheers,
Rick.