Hot Pixels when Binning

[PeterC65]

I have an IXM585 based Uranus-C camera that I use mostly for EEVA. Recently I've been experimenting with binning for observing smaller fainter objects and I've noticed what appear to be hot pixels when using binning. What I'm seeing is red and green dots appearing at random spots in the image which become wiggly lines as live stacking progresses. I think there may be blue dots / lines too but they are harder to spot. Unfortunately I don't have a snapshot.

With 2x2 binning these hot pixels can be largely removed using the SharpCap hot pixel removal feature, but with 3x3 binning they make the image unusable, and with 4x4 binning they sit part way between the 2x2 and 3x3 settings.

I've done some investigation and whether I use hardware binning or software binning, sum or average binning, doesn't seem to make a difference.

Has anyone else experienced this? With this camera / sensor or others? Does anyone have a fix?

 

[Stuart1971]

Hot pixels are normal with nearly every Astro camera available, but this is why you take dark calibration frames, so when calibrated they remove the hot pixels, so they are just not an issue….also dithering helps….

[PeterC65]

3 minutes ago, Stuart1971 said:

Hot pixels are normal with nearly every Astro camera available, but this is why you take dark calibration frames, so when calibrated they remove the hot pixels, so they are just not an issue….also dithering helps….

The hot pixels (what I am calling them) appear randomly in the image and so wouldn't be removed by taking dark frames. Perhaps they are not hot pixels (see my description in the original post). I don't get them when using 1x1 binning.

 

Edited February 11, 2023 by PeterC65

[Stuart1971]

7 minutes ago, PeterC65 said:

The hot pixels (what I am calling them) appear randomly in the image and so wouldn't be removed by taking dark frames. Perhaps they are not hot pixels (see my description in the original post). I don't get them when using 1x1 binning.

 

Yes they sound like hot pixels, and they will get worse when binning, that is normal, as you are combining pixels, so they will get progressively bigger, 2x2 bin is 4 pixels combined, so the hot pixel is 4 times bigger, and they are always random, and usually are green, red or blue, or white before debayering…

Dark frame will calibrate these out trust me, i have them with my camera and so will most people, and they look awful until they are calibrated….

Your  dark frames have to be taken as excactly the same temp and exposure duration as the lights frames to calibrate properly…

They will probably still be there on 1x1 bin, but you will have to zoom in massively to see them…

Edited February 11, 2023 by Stuart1971

[vlaiv]

7 minutes ago, PeterC65 said:

The hot pixels (what I am calling them) appear randomly in the image and so wouldn't be removed by taking dark frames. Perhaps they are not hot pixels (see my description in the original post). I don't get them when using 1x1 binning.

 

If you don't get them using 1x1 binning - then there is no reason to get them when using software binning on any level (2x2, 3x3, ...).

Software binning is preferred way of binning with modern CMOS sensors btw.

[PeterC65]

1 hour ago, vlaiv said:

If you don't get them using 1x1 binning - then there is no reason to get them when using software binning on any level (2x2, 3x3, ...).

It may be that I do get them but don't notice. They are noticeable at 2x2, very noticeable at 3x3, and at 4x4 they are more noticeable than at 2x2 but less noticeable than at 3x3, so they don't change linearly with the degree of binning.

The dots I am seeing are not fixed in position between live stacks. For a given live stack they stay in position which is why I get the wiggly lines, but when I start a new live stack they can appear in a new place. This is making me think they may not be hot pixels.

1 hour ago, vlaiv said:

Software binning is preferred way of binning with modern CMOS sensors btw.

Why is this?

With hardware binning I seem to be limited to average binning only which isn't useful when I'm trying to increase the light grab.

 

[ONIKKINEN]

I would be interested to learn just how exactly does the binning happen on camera with the software you use. If it does debayering first to reach 1x and then bins to the desired value there will have been interpolation done on the raw frame. Debayering algorithms have a habit of turning single hot pixels into larger hot pixels, because the debayering process is just a guess on what colours should go in the blank spaces between pixels so there are mistakes (normal imaging employs dithering, so not an issue) Superpixel debayering, if an option, bins x2 and has no interpolation issues.

Not debayering and binning results in a mono image, because the data from OSC cameras is really mono.

On light grasp of binned images, no difference in addition and averaging methods because its not the signal you are after but the signal to noise ratio.

[PeterC65]

32 minutes ago, ONIKKINEN said:

I would be interested to learn just how exactly does the binning happen on camera with the software you use. If it does debayering first to reach 1x and then bins to the desired value there will have been interpolation done on the raw frame. Debayering algorithms have a habit of turning single hot pixels into larger hot pixels, because the debayering process is just a guess on what colours should go in the blank spaces between pixels so there are mistakes (normal imaging employs dithering, so not an issue) Superpixel debayering, if an option, bins x2 and has no interpolation issues.

Not debayering and binning results in a mono image, because the data from OSC cameras is really mono.

On light grasp of binned images, no difference in addition and averaging methods because its not the signal you are after but the signal to noise ratio.

I have no information about how the hardware or software binning is done.

I'm using SharpCap for software binning and that does both sum and average binning, the camera hardware just seems to do average binning.

If I use sum binning then the image gets noticeably brighter, and I am after the signal because when doing EEVA I want to keep the exposures short.

SharpCap is also debayering the image.

 

[ONIKKINEN]

6 minutes ago, PeterC65 said:

I have no information about how the hardware or software binning is done.

I'm using SharpCap for software binning and that does both sum and average binning, the camera hardware just seems to do average binning.

If I use sum binning then the image gets noticeably brighter, and I am after the signal because when doing EEVA I want to keep the exposures short.

SharpCap is also debayering the image.

 

I think you kind of missed the point with signal here. How bright the image appears is surely of no concern if it contains no extra information? After all if you were after signal only you would put the camera to the maximum gain it can go and get a very bright image.

With average binning the same signal to noise ratio improvement happens than with sum binning, the image just doesn't get brighter, but you can extract more information out of it by applying a harder stretch - which sharpcap should do for you if you click the autostretch button in the histogram. Its possible you get nothing more out of it if the SNR gain is so little but this also applies to the brightened addition binned image. By the way binning with CMOS cameras does not allow you to get away with shorter exposures as read noise gets "read" from the individual pixels whether you bin or not. But of course whatever little positive SNR you have from a short exposure will be multiplied by the bin factor, so it is still very useful if you are oversampling and the extra resolution is of no value to the observations - just not something you can rely on to make very short exposures be much better.

[PeterC65]

3 minutes ago, ONIKKINEN said:

I think you kind of missed the point with signal here. How bright the image appears is surely of no concern if it contains no extra information? After all if you were after signal only you would put the camera to the maximum gain it can go and get a very bright image.

With average binning the same signal to noise ratio improvement happens than with sum binning, the image just doesn't get brighter, but you can extract more information out of it by applying a harder stretch - which sharpcap should do for you if you click the autostretch button in the histogram. Its possible you get nothing more out of it if the SNR gain is so little but this also applies to the brightened addition binned image. By the way binning with CMOS cameras does not allow you to get away with shorter exposures as read noise gets "read" from the individual pixels whether you bin or not. But of course whatever little positive SNR you have from a short exposure will be multiplied by the bin factor, so it is still very useful if you are oversampling and the extra resolution is of no value to the observations - just not something you can rely on to make very short exposures be much better.

Remember I'm doing EEVA, so what I want is to see the object in an OK way relatively quickly. I use 4s exposures almost exclusively and live stack for no more than 5 minutes. Recently I've been using a Barlow to magnify smaller objects (x2.4), but that slows the scope (F5 becomes F12), and sum binning counters that, allowing me to stick with the 4s exposure time (average binning does not).

I use histogram stretch, both autostretch and manually, but the effect depends on the starting point and with small / faint objects needing a Barlow I'm getting better results using 2x2 binning. It's just that these 'hot pixels' are getting in the way!

 

[ONIKKINEN]

2 minutes ago, PeterC65 said:

Recently I've been using a Barlow to magnify smaller objects (x2.4), but that slows the scope (F5 becomes F12), and sum binning counters that, allowing me to stick with the 4s exposure time (average binning does not).

Not sure what to say if you dont want to believe me, but this is not correct at all regarding binning. Sum binning and average binning results in the exact same SNR improvement and you are not getting anything more out of sum binning here, you are just getting a brighter image with no extra benefit in it the same way as if you just multiplied all pixels values by 4x (with 2x2 bin). I am guessing autostretch does not stretch the image further if you average bin because there is no point (low SNR).

Actually this mention of a barlow sheds some light to the issue. You are using a very small pixel camera, very short exposures and now a barlow lens that turns the scope into a very slow one, no wonder you have hot pixels - there is hardly any other signal going on at F/12 with 4s exposures!

Here is what i suggest, not strictly a direct fix to your hot pixel dilemma but will help your EEVA adventures immensely: Drop the barlow, only use it for lunar and planetary. You are really killing the efficiency of your system here with using that and there is no way for me to get this point across too urgently! I am 99% sure you get no extra actual detail with the barlow and its just pure oversampling - which results in a pure loss of SNR with no extra gain whatsoever but it gets you the closer view you wanted. Here is an easier method that does not sacrifice SNR and maybe you dont need to bin (as much, probably still want to): Set the capture area to be smaller around the object you are observing, its as simple as that really. Now the view is zoomed in, you are still at F/5 so the 4s exposures are given a much better chance at success and since you were oversampling with the barlow you actually have the exact same amount of detail in the image! No negatives this way.

[PeterC65]

4 minutes ago, ONIKKINEN said:

Not sure what to say if you dont want to believe me, but this is not correct at all regarding binning. Sum binning and average binning results in the exact same SNR improvement and you are not getting anything more out of sum binning here, you are just getting a brighter image with no extra benefit in it the same way as if you just multiplied all pixels values by 4x (with 2x2 bin). I am guessing autostretch does not stretch the image further if you average bin because there is no point (low SNR).

Actually this mention of a barlow sheds some light to the issue. You are using a very small pixel camera, very short exposures and now a barlow lens that turns the scope into a very slow one, no wonder you have hot pixels - there is hardly any other signal going on at F/12 with 4s exposures!

Here is what i suggest, not strictly a direct fix to your hot pixel dilemma but will help your EEVA adventures immensely: Drop the barlow, only use it for lunar and planetary. You are really killing the efficiency of your system here with using that and there is no way for me to get this point across too urgently! I am 99% sure you get no extra actual detail with the barlow and its just pure oversampling - which results in a pure loss of SNR with no extra gain whatsoever but it gets you the closer view you wanted. Here is an easier method that does not sacrifice SNR and maybe you dont need to bin (as much, probably still want to): Set the capture area to be smaller around the object you are observing, its as simple as that really. Now the view is zoomed in, you are still at F/5 so the 4s exposures are given a much better chance at success and since you were oversampling with the barlow you actually have the exact same amount of detail in the image! No negatives this way.

I do believe you about the SNR with sum and average binning!

That's an interesting point you made about the Barlow being the potential cause of the hot pixels. I don't remember hot pixels being a problem without the Barlow. I will check. Hopefully that will put my mind to rest about the hot pixels (I've been worried it might be a camera issue).

I've tried observing smaller objects without a Barlow. I can zoom in digitally (or use a small ROI as you suggest) but then the object rapidly pixilates. My thinking is that the Barlow magnifies the object so that it covers more pixels and therefore pixilates less. Then binning keeps the system from oversampling but I suppose it just reintroduces the pixilation issue which I suppose is another way of saying what you are saying above (there is no more information to be had).

Are we saying that to observe smaller, fainter objects I need more aperture, even with EEVA? I guess that with AP (not EEVA) it's possible to use long exposures times (totalling hours) rather than succumb to aperture fever.

 

[ONIKKINEN]

6 minutes ago, PeterC65 said:

I do believe you about the SNR with sum and average binning!

That's an interesting point you made about the Barlow being the potential cause of the hot pixels. I don't remember hot pixels being a problem without the Barlow. I will check. Hopefully that will put my mind to rest about the hot pixels (I've been worried it might be a camera issue).

I've tried observing smaller objects without a Barlow. I can zoom in digitally (or use a small ROI as you suggest) but then the object rapidly pixilates. My thinking is that the Barlow magnifies the object so that it covers more pixels and therefore pixilates less. Then binning keeps the system from oversampling but I suppose it just reintroduces the pixilation issue which I suppose is another way of saying what you are saying above (there is no more information to be had).

Are we saying that to observe smaller, fainter objects I need more aperture, even with EEVA? I guess that with AP (not EEVA) it's possible to use long exposures times (totalling hours) rather than succumb to aperture fever.

 

Regarding the barlow and bin thing reintroducing pixelation, you are exactly right. But actually its much worse now since the subs have much lower SNR before the bin happens, so binning back to the no-barlow resolution has only negatives. For fainter and smaller objects there is unfortunately no fast-lane to overtake the issues. Just need more time or a bigger scope, or both if the object is so faint. Binning improves SNR so you get a better image faster, but of course the image is now smaller. The most difficult thing to do in all of astrophotography (including EEVA i suppose) is to get a highly detailed and large dimensioned image of a faint distant target!

[vlaiv]

4 hours ago, PeterC65 said:

Why is this?

With hardware binning I seem to be limited to average binning only which isn't useful when I'm trying to increase the light grab.

Because of limited bandwidth for data download and also because you can choose binning method to suit you.

As far as data itself - in principle there is no difference in software and hardware (or rather firmware) binning with modern CMOS sensors. You should in principle get the same result - but in practice it is not so because of other limiting factors - namely data format.

Say that your camera is 12bit.

Binning by its nature increases bit depth of the data. Regardless of the binning method - average and sum are same (except for division factor - but that is constant over whole image and is not important as far as SNR goes).

Add 4 12bit numbers and you'll get 14bit number.

Average 4 12bit numbers - and you will get 12bit number with 2 bits after decimal point.

In either case - you need more bits to accurately represent number that you get from binning. However - firmware binning still produces 12bit number. It looses that precision that you gained with binning.

With software binning - you will download original 12bits - and then you can do all the operations in floating point - without any loss in precision (although binning still produces "round" number of bits - there are operations that are not like that and are therefore better represented in floating point format).

You can also choose to do "clever"-er binning.

This would greatly improve results in your case for example.

Say that you have telegraph noise like in your example. Btw - that is probably telegraph noise that you are seeing - it looks like hot pixels but is not consistent and pops up in different places much like Morse code (hence name telegraph noise - dash dot dot dash dot ....)

You can choose to bin 2x2 pixels and take average of all four - or you can choose to skip one pixel if you determine it is anomalous and just average remaining 3 that you think have proper value (that can be determined by some sort of threshold - like if one of pixels is much larger than average of the four - odds are that it is hot and should not be taken into account).

 

[PeterC65]

21 minutes ago, vlaiv said:

Binning by its nature increases bit depth of the data. Regardless of the binning method - average and sum are same (except for division factor - but that is constant over whole image and is not important as far as SNR goes).

Add 4 12bit numbers and you'll get 14bit number.

Average 4 12bit numbers - and you will get 12bit number with 2 bits after decimal point.

In either case - you need more bits to accurately represent number that you get from binning. However - firmware binning still produces 12bit number. It looses that precision that you gained with binning.

With software binning - you will download original 12bits - and then you can do all the operations in floating point - without any loss in precision (although binning still produces "round" number of bits - there are operations that are not like that and are therefore better represented in floating point format).

That makes sense. I will stick with software binning.

22 minutes ago, vlaiv said:

Say that you have telegraph noise like in your example. Btw - that is probably telegraph noise that you are seeing - it looks like hot pixels but is not consistent and pops up in different places much like Morse code (hence name telegraph noise - dash dot dot dash dot ....)

That also makes sense, thanks!

24 minutes ago, vlaiv said:

You can choose to bin 2x2 pixels and take average of all four - or you can choose to skip one pixel if you determine it is anomalous and just average remaining 3 that you think have proper value (that can be determined by some sort of threshold - like if one of pixels is much larger than average of the four - odds are that it is hot and should not be taken into account).

Can you point me at software that will do this sort of binning?

 

[vlaiv]

Just now, PeterC65 said:

Can you point me at software that will do this sort of binning?

I'm not really sure which one does it - but it is the same thing as sigma reject when stacking - with enough pixels one can determine statistics of sample and anything that falls out of statistics can be rejected.

I'm sure this can easily be implemented as PI script for example (pixel math thing). I've done it for ImageJ as plugin.

Not sure which EEVA software does it though.