Jump to content

  •  

CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.

Photo

Long Exposure Noise Reduction Vs. Dark Frame Subtraction

  • Please log in to reply
50 replies to this topic

#1 James Paulson

James Paulson

    Vanguard

  • -----
  • topic starter
  • Posts: 2,008
  • Joined: 09 Mar 2014

Posted 01 October 2018 - 06:39 PM

I attended a seminar this weekend and it was advocated to use the cameras LENR functions to shoot and subtract the dark frame using the cameras in house LENR at the time of composition as opposed to waiting until the end, shooting darks and then subtracting them within DSS.

 

What is the consensus and latest research on this in the AP community? The argument for using LENR was pretty strong.



#2 Darren in Tacoma

Darren in Tacoma

    Messenger

  • -----
  • Posts: 428
  • Joined: 11 Feb 2012

Posted 01 October 2018 - 06:54 PM

I would see how much imaging time you lose doing noise reduction during the dark hours. I personally want to capture as much as possible due to short number of dark hours in the summer and few clear nights year round.



#3 Jim Waters

Jim Waters

    Fly Me to the Moon

  • *****
  • Posts: 6,251
  • Joined: 21 Oct 2007
  • Loc: Phoenix, AZ USA

Posted 01 October 2018 - 06:57 PM

If I shoot a single wide-field Light Sub at high ISO (6400 or higher) I use LENR and process in Lightroom.  If I shoot multiple Light Subs I take Darks along with BIAS and Flats and process in PixInsight.

 

IMHO using LENR and stacking the Light Subs and processing them will produce poor results.  Its better to take Darks, BIAS and Flats.



#4 xiando

xiando

    Fly Me to the Moon

  • *****
  • Posts: 6,822
  • Joined: 27 May 2015
  • Loc: Cloudy NEOhio

Posted 01 October 2018 - 07:04 PM

If you're not using a cooled temperature controlled camera (within its control circuitry's delta T, I might add), that's probably the best option.


Edited by xiando, 01 October 2018 - 07:05 PM.

  • WarmWeatherGuy and mvas like this

#5 Alen K

Alen K

    Surveyor 1

  • *****
  • Posts: 1,729
  • Joined: 25 Nov 2009

Posted 01 October 2018 - 08:27 PM

I attended a seminar this weekend and it was advocated to use the cameras LENR functions to shoot and subtract the dark frame using the cameras in house LENR at the time of composition as opposed to waiting until the end, shooting darks and then subtracting them within DSS.

Was it a seminar by Alan Dyer? He's pretty big on that approach.

IMHO using LENR and stacking the Light Subs and processing them will produce poor results.

Can you explain?

#6 freestar8n

freestar8n

    Vendor - MetaGuide

  • *****
  • Vendors
  • Posts: 10,872
  • Joined: 12 Oct 2007
  • Loc: Melbourne, Australia

Posted 01 October 2018 - 08:34 PM

If the camera is not cooled and the temperature is changing - either due to ambient temperature changes or the camera heating up - there could well be much more benefit from having the darks be a good match to each light vie LENR - compared to the time lost imaging.  Losing light signal in a given time is bad - but what matters is the ratio of signal to noise - and better matching of the dark current pattern noise to each light could result in higher snr despite less signal fromt he object.

 

There is no way to know without characterizing the pattern noise and how it is changing.  If there is no pattern noise, or if the noise is not changing, then you would be better off not using LENR.

 

It also depends on how much you rely on cosmetic clean up in post-processing - as opposed to the quality of the data before manual manipulation.

 

Frank


  • 42itous1 likes this

#7 Jon Rista

Jon Rista

    ISS

  • *****
  • Posts: 24,988
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 01 October 2018 - 08:36 PM

LENR has its place. If you have to image with a DSLR on warmer nights, LENR could produce a better image, despite the loss in total integration. It may not necessarily have the best SNR, but image quality is not solely about SNR. Glows, banding and other FPN issues that can show up with high temperatures are often a big eye sore, and are also usually MUCH better corrected with LENR than a dark library. I recently answered this same question here, so I'll just link to the other discussion:

 

https://www.cloudyni...rary/?p=8845572

 

I have actual examples of darks corrected with less than perfectly matched other darks taken later in similar, but not exactly the same, conditions...as well as darks corrected with LENR. You can clearly see the difference, and the benefit of LENR, even at the extremely high EXIF temp of 46C. The big thing here is...we actually never know what the SENSOR temp is. EXIF temps from Canon RAW images are actually the DSP (DIGIC) processor temps, and from Sony RAW are actually the battery temps! So even if you have an EXIF temp in your lights of say 30C...and you get some more darks that also have EXIF temps of 30C, there is ZERO guarantee that you will actually have a match between the darks and the lights. A good match is often critical with higher temps to actually correct the FPN, and without a good match, it is a total tossup. Because of this, even with a meticulous dark library, the only key you really have to match on is temperature, the EXIF temperature, and if that key is effectively meaningless, or less meaningful than you think, then there is no real good way to effectively use a well constructed dark library.

 

LENR is not always useful. During winter, for example, I would say skip it and use darks taken after the fact. When dark current overall is much lower, then the discrepancy between less than perfectly matched frames doesn't matter as much. LENR's place is during warm nights where sensor temps will be high. Anyway...more in the link above (I have numerous posts in there starting with the one I linked).


  • happylimpet, JuergenB, 42itous1 and 1 other like this

#8 Jim Waters

Jim Waters

    Fly Me to the Moon

  • *****
  • Posts: 6,251
  • Joined: 21 Oct 2007
  • Loc: Phoenix, AZ USA

Posted 01 October 2018 - 08:37 PM

Was it a seminar by Alan Dyer? He's pretty big on that approach.

Can you explain?

Lets say you have 2 total of imaging time before the object gets too low or moves into light pollution.  When you take LENR you are effectively limiting yourself to 1 hour.  LENR takes an automatic Dark after each Light.  Without LENR your get the entire 2 hours.  The more Lights you take the S/N ratio will be better.  You can take Darks when you are packing up or later. 


Edited by Jim Waters, 01 October 2018 - 08:38 PM.


#9 Jon Rista

Jon Rista

    ISS

  • *****
  • Posts: 24,988
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 01 October 2018 - 08:48 PM

 The more Lights you take the S/N ratio will be better.

The key here is...this statement is not necessarily true. It depends on whether your SNR is more affected by FPN (glows, banding, temporally fixed spatially random noise due to DSNU) or not. There are cases where ideal dark correction produces a better looking image than poor dark correction, even if you reduce temporally random noise more by stacking more subs.

 

In many case, it is true that taking more lights delivers better SNR...but it is not always the case. Summer is a key time when LENR could be useful, DESPITE teh loss in sub count, because it manages the often very nasty dark signal better than even a carefully crafted dark library does after the fact.



#10 mvas

mvas

    Apollo

  • -----
  • Posts: 1,337
  • Joined: 27 May 2015
  • Loc: Eastern Ohio

Posted 01 October 2018 - 09:00 PM

Lets say you have 2 total of imaging time before the object gets too low or moves into light pollution.  When you take LENR you are effectively limiting yourself to 1 hour.  LENR takes an automatic Dark after each Light.  Without LENR your get the entire 2 hours.  The more Lights you take the S/N ratio will be better.  You can take Darks when you are packing up or later. 

How are you matching the temperature of those Lights and those Darks when the sensor temperature changing all night long?



#11 Jim Waters

Jim Waters

    Fly Me to the Moon

  • *****
  • Posts: 6,251
  • Joined: 21 Oct 2007
  • Loc: Phoenix, AZ USA

Posted 01 October 2018 - 09:55 PM

How are you matching the temperature of those Lights and those Darks when the sensor temperature changing all night long?

I don't do an exact temperature match.  I have found that with my Canon 6D if I am within +- 5 degrees I am OK.  



#12 Jim Waters

Jim Waters

    Fly Me to the Moon

  • *****
  • Posts: 6,251
  • Joined: 21 Oct 2007
  • Loc: Phoenix, AZ USA

Posted 01 October 2018 - 10:04 PM

The key here is...this statement is not necessarily true. It depends on whether your SNR is more affected by FPN (glows, banding, temporally fixed spatially random noise due to DSNU) or not. There are cases where ideal dark correction produces a better looking image than poor dark correction, even if you reduce temporally random noise more by stacking more subs.

 

In many case, it is true that taking more lights delivers better SNR...but it is not always the case. Summer is a key time when LENR could be useful, DESPITE teh loss in sub count, because it manages the often very nasty dark signal better than even a carefully crafted dark library does after the fact.

With my Ha Modified 6D I don't suffer from banding or amp glow.  I have tried using LENR and feel that for me its more effective to get as many Light Subs as I can.  I shoot at ISO 1600 and try to keep my subs under 300".  I don't know if LENR is more effective at different ISO's and exposure times.



#13 freestar8n

freestar8n

    Vendor - MetaGuide

  • *****
  • Vendors
  • Posts: 10,872
  • Joined: 12 Oct 2007
  • Loc: Melbourne, Australia

Posted 01 October 2018 - 10:28 PM

With my Ha Modified 6D I don't suffer from banding or amp glow.  I have tried using LENR and feel that for me its more effective to get as many Light Subs as I can.  I shoot at ISO 1600 and try to keep my subs under 300".  I don't know if LENR is more effective at different ISO's and exposure times.

There is no way to know a priori if one way is better or not.  It will depend on pattern noise and thermal properties of the camera that are not well known or described.

 

Theoretically if the sensor is always at some uniform temperature then you could match it to a library of darks at different temperatures - but if there is a temperature gradient across the sensor this may not work well at all - whereas taking a dark immediately after each light will capture that gradient directly - except for a small change in the elapsed time.

 

So if anyone does the comparison and finds one way works better than the other - they should stick to the better way.  But it will depend on the camera, the environment, the exposures - etc.

 

I remember imaging on a warm night with a Canon 20D - and the thermal noise was so high and changing that I'm certain in camera dark subtraction would have worked better.  But cameras have improved a lot since then.

 

Frank



#14 Jon Rista

Jon Rista

    ISS

  • *****
  • Posts: 24,988
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 02 October 2018 - 12:37 AM

With my Ha Modified 6D I don't suffer from banding or amp glow.  I have tried using LENR and feel that for me its more effective to get as many Light Subs as I can.  I shoot at ISO 1600 and try to keep my subs under 300".  I don't know if LENR is more effective at different ISO's and exposure times.

This can depend a lot on the sample of the camera. I've processed a lot of 6D data from various individuals, and some don't have quite as bad of amp glow, others have much worse. Also, how much amp glow you see may also depend on how deep your exposures are, and of course how high the temps are. A lot of the glows in DSLRs come from the heat of other components, such as the DIGIC chips in Canon DSLRs.

 

And, as I said before, LENR is not always the best solution. During summer months on warm nights, it can produce better quality images, but again, it would depend on the night in many cases.

 

Anyway, it is not as simple as simply saying more subs = better. It MAY be the case, however it depends, and there are times when LENR could be better. You would generally need to get a feel for your particular camera, especially at warmer temps (i.e. 25C ambient, where sensor temps could be 30-35C or more, especially if you use live view regularly for focusing...then you could easily have sensor temps closer to 40C or more, and THAT is where LENR has the potential to improve IQ. I have imaged on numerous nites at dark sites, with warm ambient temps, and had problems with both amp glow and just the high dark current that I could not properly calibrate out later on despite a large dark library. The difference in dark site data, even with 7-10 hours worth, with that much dark current and DFPN, compared to dark site data from winter nights, is often quite obvious. At the time, I was against LENR, and did not use it, and it was only later after doing my own experiments that I learned how valuable it could be.


  • Jim Waters and JuergenB like this

#15 James Paulson

James Paulson

    Vanguard

  • -----
  • topic starter
  • Posts: 2,008
  • Joined: 09 Mar 2014

Posted 02 October 2018 - 06:59 AM

There is no way to know a priori if one way is better or not.  It will depend on pattern noise and thermal properties of the camera that are not well known or described.

 

Theoretically if the sensor is always at some uniform temperature then you could match it to a library of darks at different temperatures - but if there is a temperature gradient across the sensor this may not work well at all - whereas taking a dark immediately after each light will capture that gradient directly - except for a small change in the elapsed time.

 

So if anyone does the comparison and finds one way works better than the other - they should stick to the better way.  But it will depend on the camera, the environment, the exposures - etc.

 

I remember imaging on a warm night with a Canon 20D - and the thermal noise was so high and changing that I'm certain in camera dark subtraction would have worked better.  But cameras have improved a lot since then.

 

Frank

So have cameras improved a lot since then? These are measurable and real things.



#16 RedLionNJ

RedLionNJ

    Aurora

  • *****
  • Posts: 4,706
  • Joined: 29 Dec 2009
  • Loc: Red Lion, NJ, USA

Posted 02 October 2018 - 07:46 AM

"Signal-to-noise" ratio doesn't just apply to lights - it applies to all types of frames (darks, flats, bias).  "Noise" has a different meaning in the case of darks, but it still comes with a "fixed" element and a random element. By averaging a ton of darks, we improve the ratio of fixed to random - and that's a pretty good reason for not using in-camera NR.  So we build a master dark (and masters of all the other calibration frame types) instead.  Then we subtract that master dark from each light, etc.

 

There's no doubt temperature-matching matters. But by and large, I've found the reported (processor) temperature to have a sufficient-enough correlation to the noise in my frames to be able to consider the reported temperature good enough for temperature-matching purposes, within maybe 2 degrees C.

 

And (as mentioned by a few, above) - with EXTREMELY limited dark, clear skies with good seeing - light exposure time is king. I can do darks later, or re-use some from my library (provided they're relatively recent).



#17 freestar8n

freestar8n

    Vendor - MetaGuide

  • *****
  • Vendors
  • Posts: 10,872
  • Joined: 12 Oct 2007
  • Loc: Melbourne, Australia

Posted 02 October 2018 - 03:59 PM

So have cameras improved a lot since then? These are measurable and real things.

Yes they have improved a lot.  But the critical factors are the pattern noise and its consistency over time and with temperature changing.  Those are very real things but are not spec'd - and are hard to quantify.

 

Most discussions of noise and SNR in imaging just look at shot noise and its reduction in a stack.  But that completely misses the main noise terms that can kill the SNR even though S is increasing linearly in the stack.

 

Frank


  • Susanna144 likes this

#18 WarmWeatherGuy

WarmWeatherGuy

    Aurora

  • *****
  • Posts: 4,696
  • Joined: 27 Aug 2011
  • Loc: Las Cruces, NM 32° N, 107° W

Posted 02 October 2018 - 04:09 PM

I took 86 darks of 20 seconds each over a half hour at ISO 6400 using my Lumix G7. Apparently the sensor heated up during that time. Here is the first and 86th dark frame.

 

P1280363_s.jpg

 

 

P1280448_s.jpg


  • Susanna144 likes this

#19 Jerry Lodriguss

Jerry Lodriguss

    Vendor

  • ****-
  • Vendors
  • Posts: 7,797
  • Joined: 19 Jul 2008
  • Loc: Voorhees, NJ

Posted 02 October 2018 - 04:18 PM

So, if the temperature inside a DSLR is constantly increasing until it reaches some maximum based on the ambient temp...

 

How do any of the in-camera LENR darks match the temp of the lights, since they are not taken concurrently?

 

ICLENR darks taken before equilibrium in the camera don't match their lights cause they are not taken at the same time.

 

How is this different than darks on a cloudy night? Well, you say, they are probably closer. But they still don't match exactly and you are in the same boat as cloudy nights darks. You really have no idea what the temp of the light is compared to the temp of the ICLENR dark.  All I think you can say is the ICLENR dark is going to be hotter than the light it is subtracted from.

 

You and I don't really know the sensor temp from the EXIF data. But whatever it is reading is what you have to work with.  I don't think it is too wrong of an assumption to believe that whatever temp it is reading is going to be a consistent relationship to the actual sensor temp. So for all practical purposes, the EXIF temp is usable.

 

If you don't know the actual sensor temp, but you do know the EXIF temp, you can still figure that darks with the same EXIF temp as the lights are going to be the closest match you can get.

 

Then you get both the benefit of more signal from more lights, as well as the benefit of less calibration noise with a good master dark shot on a cloudy night in your garage.

 

LENR doesn't give you either of these benefits, plus the temp of the LENR dark is not going to exactly match the light anyway.

 

Personally, I believe that once you really believe that this stuff is critical, and you want to push the envelope to get the absolute best you can, then it's time to quit fooling around with a DSLR and get a regulated cooled astro-specific CMOS or CCD camera.

 

Otherwise, just shoot your darks on a cloudy night in your garage. And as the temp changes through the season, you end up with a library of darks about every 5 degrees or so.  Since you are shooting them in your garage on a cloudy night, you can shoot hundreds of them to make a really good master dark.

 

Then you use bias frames and some program that does automatic dark frame matching, and it will scale your darks, and you will be doing about as good as you can with a DSLR.

 

Now, if I was shooting single frames, like Starscapes, I would be using LENR. That is a different case than deepsky.

 

Jerry


Edited by Jerry Lodriguss, 02 October 2018 - 04:24 PM.

  • Jim Waters and Dom_Qc like this

#20 freestar8n

freestar8n

    Vendor - MetaGuide

  • *****
  • Vendors
  • Posts: 10,872
  • Joined: 12 Oct 2007
  • Loc: Melbourne, Australia

Posted 02 October 2018 - 04:36 PM

So, if the temperature inside a DSLR is constantly increasing until it reaches some maximum based on the ambient temp...

 

How do any of the in-camera LENR darks match the temp of the lights, since they are not taken concurrently?

 

ICLENR darks taken before equilibrium in the camera don't match their lights cause they are not taken at the same time.

 

How is this different than darks on a cloudy night? Well, you say, they are probably closer. But they still don't match exactly and you are in the same boat as cloudy nights darks. You really have no idea what the temp of the light is compared to the temp of the ICLENR dark.  All I think you can say is the ICLENR dark is going to be hotter than the light it is subtracted from.

 

You and I don't really know the sensor temp from the EXIF data. But whatever it is reading is what you have to work with.  I don't think it is too wrong of an assumption to believe that whatever temp it is reading is going to be a consistent relationship to the actual sensor temp. So for all practical purposes, the EXIF temp is usable.

 

If you don't know the actual sensor temp, but you do know the EXIF temp, you can still figure that darks with the same EXIF temp as the lights are going to be the closest match you can get.

 

Then you get both the benefit of more signal from more lights, as well as the benefit of less calibration noise with a good master dark shot on a cloudy night in your garage.

 

LENR doesn't give you either of these benefits, plus the temp of the LENR dark is not going to exactly match the light anyway.

 

Personally, I believe that once you really believe that this stuff is critical, and you want to push the envelope to get the absolute best you can, then it's time to quit fooling around with a DSLR and get a regulated cooled astro-specific CMOS or CCD camera.

 

Otherwise, just shoot your darks on a cloudy night in your garage. And as the temp changes through the season, you end up with a library of darks about every 5 degrees or so.  Since you are shooting them in your garage on a cloudy night, you can shoot hundreds of them to make a really good master dark.

 

Then you use bias frames and some program that does automatic dark frame matching, and it will scale your darks, and you will be doing about as good as you can with a DSLR.

 

Now, if I was shooting single frames, like Starscapes, I would be using LENR. That is a different case than deepsky.

 

Jerry

There will be some error since the dark is taken some time after the light - and I allude to that above.

 

But what matters is the net error in the final result - along with the signal.

 

If there is a dynamic gradient in the sensor temperature then LENR may do a much better job of removing it from each light, compared to a master dark.

 

There is no clear answer - but when the sensor behavior is changing over time - and for many people they can see that it is directly from the appearance of their lights - there is good reason to expect LENR to be worth the price of reduced exposure time.

 

Frank



#21 Jon Rista

Jon Rista

    ISS

  • *****
  • Posts: 24,988
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 02 October 2018 - 05:09 PM

 

Personally, I believe that once you really believe that this stuff is critical, and you want to push the envelope to get the absolute best you can, then it's time to quit fooling around with a DSLR and get a regulated cooled astro-specific CMOS or CCD camera.

 

waytogo.gif

 

This right here really is the right answer to the dark current problem. DSLRs are great, until they are not... 


  • WarmWeatherGuy likes this

#22 Alen K

Alen K

    Surveyor 1

  • *****
  • Posts: 1,729
  • Joined: 25 Nov 2009

Posted 02 October 2018 - 06:05 PM

So, if the temperature inside a DSLR is constantly increasing until it reaches some maximum based on the ambient temp...
 
How do any of the in-camera LENR darks match the temp of the lights, since they are not taken concurrently?

It's possible that the engineers of the camera know what happens to the temperature when it takes a LENR dark frame and compensate accordingly. The camera's processor probably has access to data that isn't provided to users. (There was certainly lots of that in the products I helped design; not cameras.) We just don't know. There is a lot of faith involved with LENR. smile.gif

#23 Jon Rista

Jon Rista

    ISS

  • *****
  • Posts: 24,988
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 02 October 2018 - 06:23 PM

It's possible that the engineers of the camera know what happens to the temperature when it takes a LENR dark frame and compensate accordingly. The camera's processor probably has access to data that isn't provided to users. (There was certainly lots of that in the products I helped design; not cameras.) We just don't know. There is a lot of faith involved with LENR. smile.gif

In my experience, LENR frames calibrate the prior frame about as perfectly as possible unless the temps are EXTREMELY high (i.e. over 45C, which is very hot for an imaging sensor!), at which point even with LENR, the rate of cooling vs, heating of the sensor, and the causes of either, result in more rapid changes (i.e. you might get a light frame at 45C and the LENR frame might hit 48C, then the next light and LENR frame both drop to 40C, etc.) At such high temps cooling can very rapidly cool the sensor, even with just the passive cooling most DSLRs have. 

 

That said, under 40C, and every LENR frame I've ever tested appears to be, effectively, perfect. In dark-only tests, they usually look like this:

 

uD622vc.jpg



#24 freestar8n

freestar8n

    Vendor - MetaGuide

  • *****
  • Vendors
  • Posts: 10,872
  • Joined: 12 Oct 2007
  • Loc: Melbourne, Australia

Posted 03 October 2018 - 06:11 AM

You could also do LENR manually by just taking light/dark/light/dark and subtract them in pairs outside of the camera.  That way you see the direct benefit of darks that are a better match to lights - without relying on some other kind of funny business going on in the firmware to make the result appear cleaner.

 

If the dark current is changing with each frame then I would expect the manual approach to have clear benefit - especially if there are dynamic gradients involved that don't change too much between the light and the dark.

 

You know that if you have changing gradients in the dark - there is no way a single dark master could work well with all the exposures.

 

It might be possible to model the thermal gradient in each light and scale the temperature locally with a single master dark - and maybe some software out there tries to do that.   But if each light is individually subtracting off a good measure of the current state of the dark current - that's a big win in noise reduction.

 

And as usual - yes when you do that subtraction the shot noise will increase in the result.  But the main reason for doing the subtraction is to remove the variable dark current across the light - and that variation in dark current is a component of noise in the image.  Subtracting the dark will reduce the total noise despite the increase in another, smaller component - which is shot noise.

 

Frank



#25 Fred76

Fred76

    Ranger 4

  • -----
  • Posts: 379
  • Joined: 25 Nov 2008
  • Loc: Normandy, France

Posted 03 October 2018 - 06:22 AM

Usually, astronomical post processing softwares match the darks to the temperature (in fact noise) of the lights so that a slight temperature deviation is not detrimental to the image quality. This is possible because the thermal noise is a function of the temperature.

 

Therefore, it is not a problem if you shot your darks after you have shot your lights. During warm nights it may be worth shooting intermediate batches of darks (for example take 20 lights, then take 10 darks, and so on).

 

However, if you shoot with LENR ON, never substract bias ! The bias will be contained with thermal noise inside each individual images. Just apply the flatfield correction. When you will stack your pile of LENR-lights, you will average the noises all together. 




CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.


Recent Topics






Cloudy Nights LLC
Cloudy Nights Sponsor: Astronomics