Night Photography Reducing Noise

[Stagecoach]

On another forum I follow someone posted a method of reducing noise in foregrounds of night shots so I had a play and thought I'd share my findings.

If we imagine a night scene out in the wilderness part foreground and part sky (maybe stars or milky way) there are times when the foreground scene to skyline is very dark and it's not practical to illuminate it further. We can of course bump up the ISO and give it a long exposure but it may be too noisy for us.

It was presented that a way to address this in camera is to use a low ISO and take a multi exposure of say 6 shots whose brightness is summed giving one image of required brightness. This is done simply by turning Auto Gain OFF when taking the multi exposure. Note, this can also be done in PS with multiple images.

I had a go and found it does reduce noise considerably, but, at the expense of having an overall longer exposure time, that may or may not be detrimental.

In addition another method is to undertake a multi exposure at a higher ISO with Auto Gain ON which does not sum the brightness.

Here's the results from a trial tonight, it was dark with no moon up, all foliage was black to the eye and I could just make out the dim top of the mast. Each image was taken so that the brightness (exposure) was as equal as possible, only PP was WB. These are severe crops.



My conclusion, well it would depend upon the times needed.

Grahame

[DanK]

Interesting. Thanks for posting.

I don't know what "auto gain" is.

You wrote:

We can of course bump up the ISO and give it a long exposure but it may be too noisy for us.

Some people reading your post might not understand the difference between random (shot and read) noise and fixed pixel noise because one doesn't often encounter the latter in daytime photography. But in night photography, it's important to distinguish between them.

When people talk about noise without qualification, they almost always mean random noise. Random noise becomes more apparent at high ISOs because noise is amplified along with signal. However, it does not become more apparent with long exposures. Long exposures may generate fixed pixel noise--certain pixels that discharge inappropriately after a long time. For that reason, night photographers often use what is sometimes called long-exposure noise reduction or black-frame noise reduction: the camera takes a black frame of equal length, and since the aberrant pixels are fixed, the camera simply subtracts those pixels in from the first image.

I think I have posted this image before, but it is a good example. It is a 7-minute single exposure with black-frame noise reduction but no other noise reduction at all. You can see that it is very smooth. The long exposure didn't cause a problem with random noise.



Random noise becomes an issue in night photography under a few circumstances. One is when people boost ISO rather than using a longer exposure. (I rarely go above ISO 200 in doing night photography.) A second is if they underexpose. A photographer may have a dark image in mind, but exposing to the left generates noise; the better approach is to expose to the right and darken in post.

A third situation is the one Grahame mentioned: you have areas of very different brightness and expose for the brighter. If the bright areas aren't moving (e.g., not a rising full moon), you can merge multiple exposures as Grahame did, or you can bracket exposures and merge them in post. I've always done bracketing instead of merging, but I have never done and A/B comparison. Grahame, do you think this approach would have advantages over merging bracketed exposures in post?

[rpcrowe]

Graham...

Great post. I have found that using localized noise reduction in NIK Software, Dfine2, to assist in reduction of noise,,,

[Stagecoach]

Interesting. Thanks for posting.

I don't know what "auto gain" is.

When taking Multi exposures (on Nikon) apart from the number of exposures selectable, 2 to 10, it can deal with the merging in two ways;

a) Auto Gain OFF : Each exposure is added in a way that the final image is 'brighter' than what would have resulted from a single exposure at the same settings, A,S and ISO.

b) Auto Gain ON : Each exposure is added in a way that the final image is 'the same brightness' than what would have resulted from a single exposure at the same settings, A,S and ISO.

Some people reading your post might not understand the difference between random (shot and read) noise and fixed pixel noise because one doesn't often encounter the latter in daytime photography. But in night photography, it's important to distinguish between them.

When people talk about noise without qualification, they almost always mean random noise. Random noise becomes more apparent at high ISOs because noise is amplified along with signal. However, it does not become more apparent with long exposures. Long exposures may generate fixed pixel noise--certain pixels that discharge inappropriately after a long time. For that reason, night photographers often use what is sometimes called long-exposure noise reduction or black-frame noise reduction: the camera takes a black frame of equal length, and since the aberrant pixels are fixed, the camera simply subtracts those pixels in from the first image.

Thanks for providing others with the additional info regarding noise types.

I've always done bracketing instead of merging, but I have never done and A/B comparison. Grahame, do you think this approach would have advantages over merging bracketed exposures in post?

For clarification, the noise comparison testing was totally with respect to the foreground so consideration of any 'movement' would be such things as wind affect or moving lights. I have never undertaken bracketing with respect to assisting noise reduction but have read of people doing this. I'm wondering now if this can be accomplished in camera with multi exposure for another comparison.

What I did forget to do was to take a long exposure near base ISO to include it with the other 4.

[Stagecoach]

Graham...

Great post. I have found that using localized noise reduction in NIK Software, Dfine2, to assist in reduction of noise,,,

I too use Define for noise reduction Richard, the only concern being that any reduction generally reduces sharpness.

[DanK]

I don’t use Nikon, so this is all new to me. Ali’s the camera producing a JPEG?


Sent from my iPhone using Tapatalk

[Stagecoach]

I don’t use Nikon, so this is all new to me. Ali’s the camera producing a JPEG?


Sent from my iPhone using Tapatalk

It produces a RAW file Dan.

[DanK]

Thanks. Canon cameras, at least of the generation of mine, have no capability like that.


Sent from my iPhone using Tapatalk

[Manfred M]

I seem to remember that this can be accomplished in Photoshop.

If I remember correctly, the Scripts -> Statistics -> Mean combines different files and eliminates any data that is not found in all the images and would remove the noise. I haven't used this functionality in quite some time and don't have any files to confirm this with available to me right now.

Adobe covers off the technique here:

https://helpx.adobe.com/in/photoshop...ge-stacks.html

[Stagecoach]

I seem to remember that this can be accomplished in Photoshop.

If I remember correctly, the Scripts -> Statistics -> Mean combines different files and eliminates any data that is not found in all the images and would remove the noise. I haven't used this functionality in quite some time and don't have any files to confirm this with available to me right now.

Adobe covers off the technique here:

https://helpx.adobe.com/in/photoshop...ge-stacks.html

Manfred, reading the link it does not appear to suggest that it can 'add' the individual exposures as is done in No 3 and 4 where a low ISO was used of the comparisons, but I suspect it's possible some how in PS. But I'm aware PS can stack them similar to what was done in No 2, but it's something I have never tried.

So tonight I'll take the same scene, one single exp at high ISO and then 10 at high ISO, ten at low ISO and play with PS

[Manfred M]

Manfred, reading the link it does not appear to suggest that it can 'add' the individual exposures as is done in No 3 and 4 where a low ISO was used of the comparisons, but I suspect it's possible some how in PS. But I'm aware PS can stack them similar to what was done in No 2, but it's something I have never tried.

So tonight I'll take the same scene, one single exp at high ISO and then 10 at high ISO, ten at low ISO and play with PS

One of the members of my photo club did a talk on long exposure work / star trails just before the COVID-19 shutdown. This is how he did his work, rather than the more traditional method of long exposure. His images were remarkably noise-free.

[Stagecoach]

One of the members of my photo club did a talk on long exposure work / star trails just before the COVID-19 shutdown. This is how he did his work, rather than the more traditional method of long exposure. His images were remarkably noise-free.

Did he use the exposure 'additive' method where you end up with a single image brighter than that of one of the single exposures used, or, the 'summing' method where you end up with a single image of the same brightness than that of one of the exposures used?

[Manfred M]

Did he use the exposure 'additive' method where you end up with a single image brighter than that of one of the single exposures used, or, the 'summing' method where you end up with a single image of the same brightness than that of one of the exposures used?

Additive - the original exposures looked quite dark.

The times I used the technique was to remove people from a landmark scene and used the "same brightness" approach you used. As long as I had enough shots where I had a number of shots with people in different places, the software just eliminated them. There were a few areas where manual correction was required.

[Stagecoach]

Additive - the original exposures looked quite dark.

The times I used the technique was to remove people from a landmark scene and used the "same brightness" approach you used. As long as I had enough shots where I had a number of shots with people in different places, the software just eliminated them. There were a few areas where manual correction was required.

Using the link you gave in post 9 Manfred I have now successfully stacked a number of frames in 'Mean' mode in PS, each individual frame at the required brightness of the final merge, to considerably reduce random noise. I've not done a comparison with the 'in camera' procedure for this using 'multi exposure' but I suspect the results are similar.

I then attempted the additive mode in PS which is called 'Summation', the method where a number of 'darker' frames are 'added' to produce a final 'brighter' image as can also be done in camera. The result produced although about the brightness I wanted was rather mushy which I'm presuming was due to the individual frames being too underexposed. More experimentation is required here and I'm suspecting there will be a formula for best input exposures.

After a lot of searching I did manage to find this link that explains the procedures quite well, although I'm quite shocked at his example in scenario 2.

https://www.media-division.com/photo...-image-stacks/

[Manfred M]

Scenario 2 does look a bit suspicious, but once the authorities let me out of my cage (i.e. no longer stuck around the house) and I get a chance to get out to some more interesting scenery I'm planning to try this some more.

All of my experiments have been with Scenario 3 and it works quite effectively. That being said, during the COVID-19 "situation" having to deal with large crowds at popular locations is not going to happen soon.

[Stagecoach]

Scenario 2 does look a bit suspicious, but once the authorities let me out of my cage (i.e. no longer stuck around the house) and I get a chance to get out to some more interesting scenery I'm planning to try this some more.

All of my experiments have been with Scenario 3 and it works quite effectively. That being said, during the COVID-19 "situation" having to deal with large crowds at popular locations is not going to happen soon.

Fortunately for us there's no real restriction on daytime movement now although they say it 'should' be for essentials such as shopping but we do have a curfew from 10 pm till 5 am.

For the purposes of testing I'm lucky that I have a reasonable scene from far off city, a transmission mast and small house on a high hill to almost light free tree covered mountains to shoot from my house.

[DanK]

If I remember correctly, the Scripts -> Statistics -> Mean combines different files and eliminates any data that is not found in all the images and would remove the noise.

I don't think that's correct. I think the Mean option simply takes the average, and with enough frames, that will make the noise not apparent. Median does remove pixels that differ from the median value.

Leaving aside astrophotography, I'm still puzzled as to why one would do this rather than simply lengthening exposure as a way of avoiding random noise. One can't lengthen exposure if the subject is moving, but using the mean across images wouldn't work under those circumstances either. The only advantage I can see is that the shorter exposures would avoid fixed pixel noise, but with many modern cameras, one needs a very long exposure to create appreciable fixed noise, and it is trivial to remove it with black-frame noise reduction.

What am I missing? Are there other reasons to choose this approach for noise reduction?

[Manfred M]

Leaving aside astrophotography, I'm still puzzled as to why one would do this rather than simply lengthening exposure as a way of avoiding random noise

One explanation I've heard is that a significant source of noise in long exposure photography comes from a "hot" sensor. Cycling the sensor off and on apparently helps control that.

I have not tested that myself in a traditional way, but I have one set of images where I used a series of images to build up a night shot and another one done long exposure and I definitely noticed less noise in the multiple multiple shot version. I did not control the variables well enough to say that this is what made the difference.

[xpatUSA]

Leaving aside astrophotography, I'm still puzzled as to why one would do this rather than simply lengthening exposure as a way of avoiding random noise.

Not sure I get that, Dan. For example:

0.6sec:


25sec:


Much as Manfred suggested while I was typing.

[edit]

... the shorter exposures would avoid fixed pixel noise, but with many modern cameras, one needs a very long exposure to create appreciable fixed noise

If the reference is to fixed-pattern noise or to stuck pixels, that doesn't sound right either. [/edit]

[DanK]

One explanation I've heard is that a significant source of noise in long exposure photography comes from a "hot" sensor. Cycling the sensor off and on apparently helps control that.

I have not tested that myself in a traditional way, but I have one set of images where I used a series of images to build up a night shot and another one done long exposure and I definitely noticed less noise in the multiple multiple shot version. I did not control the variables well enough to say that this is what made the difference.

Maybe with some old and small sensors (the issue of size being the sensor box), but I am skeptical about this with reasonably modern cameras. I did have some overheating with my old 50D with very long exposures (I think over 10 minutes) on a hot, humid night, but that is a notoriously noisy camera with a small sensor box. I have never had problems with noise from overheating with my 5D III. Keep in mind that most modern sensors are fine for video recording, which keeps the sensor live for long periods, and it is always live on mirrorless cameras, if I am not mistaken. Also, it's hard to see how turning it off for a very brief moment would have much impact.


I went back and checked the editing of the 7 minute exposure I posted above. There may have been black-frame NR--one can't tell from the file, and I probably did that given the length of the exposure--but that has no impact on random noise, and I used no noise reduction of the conventional sort. Here is a 100% crop from a smooth area of that image that would show noise. It's pretty smooth.



However, it's an empirical question. I may test this at some point. One could do it without waiting for the darkest part of night by using a dense ND filter.

Ted, re your image: I suspect you got that effect because of the specifics of the camera. While I don't do many exposures as long as this example, I do have exposures frequently that are in the range of your example when I do urban night photography, and I generally get no appreciable noise and almost never use any noise reduction. For example, this one is 20 seconds: