Noise

[Rufus]

It is said that modern digital cameras are capable of making an image with far less noise than older ones.

I believe that increasing the ISO increases the sensitivity of the sensor and adversely affects the noise in the signal to noise ratio.

If I take a photograph at say ISO 1600, would one expect the noise to be similar for any combination of aperture and shutter speed that gives a consant exposure value and a central meter reading? (Assume that the shutter speed is less than 1 second to avoid complications of the sensor heating up.)

Would it be a good idea to always select the noise reduction option from the camera menu when shooting at an ISO higher than, say, 400?

The following image was taken at ISO 1600 and I feel the noise is evident in the dark sky when printed or magnified a little on screen:

[Tronhard]

It's always hard to look at a photo for such things as (subtle) noise in an image posted on-line. I can see what could be noise in the very top part of the roof particularly, but frankly I could put a case that it is also the texture of the stone or concrete...

Personally I hold the auto ISO of my gear down to 1,000 or less, although I am more allowing for monochrome images where noise (or grain if you will) can be part of the effect.

I have seen arguments for and against the proposition of increased noise with pixel density, but I think it is also a factor in sensor size and thus pixel size. For example to me, cramming 20MP into a 1/2.3" sensor is challenging to keep noise down in all but the most favourable lighting conditions.

The on-going race for Pixel size drives many people to buy one camera over another. As I have often said, a lot depends on what you are going to do with the image. Thus producing images for social media, where the image is likely going to be seen on a smaller screen and often downsized does not require massive pixel capacity.

Apologies for those who may have seen this before but it makes the point. Taken with the venerable Canon D30, a 3.2MP camera released in 2000 that had the first CMOS APS-C sensor in a DSLR. These were taken in a rather dimly lit pumping station - ,I think for social media posts they are perfectly acceptable, and at an investment for the body of $30. I can see very limited noise in them.

Both shots were taken hand-held, in available light. They were taken as RAW and lightly processed in Photoshop.



Inside the main fly wheel of a pump house 17mm, f5.6, 1/8sec, ISO-400


The fly wheel from the other side: 17mm, f5.6, 1/8sec, ISO-400


"All the beauty of life is made up of light and shadow", Leo Tolstoy;
"Skill in photography is acquired by practice and not by purchase" Percy W. Harris
"I have never been able to enlarge a photograph... I am just interested in the shots" Henri Cartier-Bresson

[Manfred M]

Actually David, the camera sensor has a single ISO (sensitivity) and what we see as ISO is really just ever increasing amplification of that signal. The more noise in the signal, the more it will be impacted by the gain from the amplification process.

The reason that you notice more digital noise in the dark areas (like the sky) is that pure black = 0 data, so any noise in the signal will show up as noise in dark, shadow areas. In places where there is very little signal, the noise, which is evenly distributed will just be much more apparent. I also believe that the blue channel is inherently the noisiest of the three channels and I don't remember why that is. I remember reading an article on the "why" and if I remember it was due to quantum effects; both the math and physics described in the article were well above my level of understanding.

[William W]

. . . Would it be a good idea to always select the noise reduction option from the camera menu when shooting at an ISO higher than, say, 400?

I assume you're referring to "Long Exposure Noise Reduction", if so, I think that it is not a good idea for me to leave it engaged as my default and it is better for me to engage it when I am shooting long exposures AND -

> I have the time to wait for the second exposure
> I am not intending HDRI / Blending / Focus Stacking or similar
> I am not making Time-lapse
> I am not intending capturing light trails or similar
> I think the image is a possibility for critical display

The following image was taken at ISO 1600 and I feel the noise is evident [ . . ] when printed or magnified a little on screen . . .

"when printed or magnified a little on screen"

I'd suggest that you do not assume the results will be the same, assuming a similar enlargement magnification and viewing distance: prints (as a general comment) will exhibit less apparent noise; and within those results different printing techniques will harvest different results.

WW

[Rufus]

It's always hard to look at a photo for such things as (subtle) noise in an image posted on-line. I can see what could be noise in the very top part of the roof particularly, but frankly I could put a case that it is also the texture of the stone or concrete...

Personally I hold the auto ISO of my gear down to 1,000 or less, although I am more allowing for monochrome images where noise (or grain if you will) can be part of the effect.

I have seen arguments for and against the proposition of increased noise with pixel density, but I think it is also a factor in sensor size and thus pixel size. For example to me, cramming 20MP into a 1/2.3" sensor is challenging to keep noise down in all but the most favourable lighting conditions.

The on-going race for Pixel size drives many people to buy one camera over another. As I have often said, a lot depends on what you are going to do with the image. Thus producing images for social media, where the image is likely going to be seen on a smaller screen and often downsized does not require massive pixel capacity.

Apologies for those who may have seen this before but it makes the point. Taken with the venerable Canon D30, a 3.2MP camera released in 2000 that had the first CMOS APS-C sensor in a DSLR. These were taken in a rather dimly lit pumping station - ,I think for social media posts they are perfectly acceptable, and at an investment for the body of $30. I can see very limited noise in them.

Both shots were taken hand-held, in available light. They were taken as RAW and lightly processed in Photoshop.

Thank you for your comment. Those are excellent images, Trev, and make your point about the number of pixels on a sensor. Although dimly lit, there are few very dark areas in them, unlike the sky in my example where the noise is particularly apparent.

My previous camera, a Canon 450D (2007) had a maximum ISO of 1600 which produced significant amounts of noise at that level. That was one reason for buying a modern camera.

[Rufus]

Actually David, the camera sensor has a single ISO (sensitivity) and what we see as ISO is really just ever increasing amplification of that signal. The more noise in the signal, the more it will be impacted by the gain from the amplification process.

Thank you for explaining the terminology more precisely. I should have said amplification of the signal from the sensor rather than sensitivity of the sensor.

The reason that you notice more digital noise in the dark areas (like the sky) is that pure black = 0 data, so any noise in the signal will show up as noise in dark, shadow areas. In places where there is very little signal, the noise, which is evenly distributed will just be much more apparent.

That nicely explains why the darker areas are more susceptible to noise at higher ISO.

[Rufus]

I assume you're referring to "Long Exposure Noise Reduction.."

My Canon 6D Mk2 has two types of noise reduction ("NR"): High ISO Speed NR and Long Exposure NR. I was thinking mainly about the first one as the second applies to exposures of 1 second or longer.

I assume you're referring to "Long Exposure Noise Reduction", if so, I think that it is not a good idea for me to leave it engaged as my default and it is better for me to engage it when I am shooting long exposures AND -

> I have the time to wait for the second exposure
> I am not intending HDRI / Blending / Focus Stacking or similar
> I am not making Time-lapse
> I am not intending capturing light trails or similar
> I think the image is a possibility for critical display

Thank you for some practical examples as the user manual gives only generic examples such as when there is subject movement, camera shake, changes in light intensity etc. It mentions these in the context of High Speed ISO NR (with Multi Shot NR enabled) but not in respect of Long Exposure NR.

The manual also says that Mutli Shot High Speed NR cannot be used with RAW or RAW+JPEG shooting, presumably because it processes 4 shots and merges them in camera. My default is to shoot in RAW, but that still leaves me 3 levels of NR without using the built in multi shot option.

[William W]

My Canon 6D Mk2 has two types of noise reduction ("NR"): High ISO Speed NR and Long Exposure NR. I was thinking mainly about the first one [i.e. "High Speed Noise Reduction"] as the second applies to exposures of 1 second or longer.

OK. Understood.

***

. . . My default is to shoot in RAW, but that still leaves me 3 levels of NR without using the built in multi shot option.

These two points might be useful for you:

1. All the information that I have gathered, confirms that High ISO Noise Reduction does NOT have an effect upon the raw file, except for embedding a metadata flag which the Canon raw file Converter, DPP, reads and thereupon acts when it converts the raw file. Third party raw converters are not necessarily affected by by this flag. It is widely assumed that none are.

2. "Long Exposure Noise Reduction" does change the raw data.

***

My opinion -

Since I always capture "raw + JPEG (L)" and I have access to both DPP and other raw file converters, then, prima facie, I think that to set "High ISO Noise Reduction" as "ON" and for that to be my default, would cause no harm; moreover, doing so would allow me a wider choice, later (i.e. not thinking about it in the field), in the Post Production phase of my workflow.

At this time I don't do that, but this conversation is rapidly heading me to that choice.

Thank you for bringing up this topic.

WW

PS - Canon User/Instruction Manuals are: Tedious, Tiresome and Tenuous.
Alliteration intended, for Tri-fold Emphasis.

[pschlute]

My default is to shoot in RAW, but that still leaves me 3 levels of NR without using the built in multi shot option.

What raw processor do you use and what are its noise reduction tools capable of doing ?

As an alternative I find selecting the black sky and bumping up the black level gets rid of noise better in situations like these.

[revi]

Thank you for explaining the terminology more precisely. I should have said amplification of the signal from the sensor rather than sensitivity of the sensor.



That nicely explains why the darker areas are more susceptible to noise at higher ISO.

Something else that makes noise more visible in dark areas:
the standard error in the signal is equal to the square root of the signal (when counting the photons).

That means that if you capture on average 100 photons/pixel over a large number of pixels, 95 % of the measured values will be between 80 and 120, i.e. +/- 20%.
But if you capture 10000 photons/pixel over a large number of pixels, you will have 95% of your pixels between 9800 and 10200, or +/- 2%...

So while the absolute noise is much higher at the higher intensity, the relative noise is much lower, so less perceptible.

Another factor that increases the perceived noise for the blue and red channels is that those two channels are amplified relative to the green channel in the white balance correction...

[Rufus]

These two points might be useful for you:

1. All the information that I have gathered, confirms that High ISO Noise Reduction does NOT have an effect upon the raw file, except for embedding a metadata flag which the Canon raw file Converter, DPP, reads and thereupon acts when it converts the raw file. Third party raw converters are not necessarily affected by by this flag. It is widely assumed that none are.

I now use the Lightroom RAW converter and previoulsy the Photoshop one.

My particular camera manual says that if High ISO NR is selected it is applied at all ISO speeds but particularly effective at a high ISO. It does not mention that DPP is required to benefit from the High ISO NR.

I agree the manual is tedious and tiresome, but nevertheless beneficial.

2. "Long Exposure Noise Reduction" does change the raw data.

I am not as widely read as you, but I would come to the same conclusion because the manual makes no reference to the possible combinations of RAW and Jpeg files, so for noise reduction to work it must alter the RAW file.

My opinion -

Since I always capture "raw + JPEG (L)" and I have access to both DPP and other raw file converters, then, prima facie, I think that to set "High ISO Noise Reduction" as "ON" and for that to be my default, would cause no harm; moreover, doing so would allow me a wider choice, later (i.e. not thinking about it in the field), in the Post Production phase of my workflow.

At this time I don't do that, but this conversation is rapidly heading me to that choice.

Your camera may differ but mine definitely does not permit High ISO NR with Multi Shot NR when the file type is set for RAW+Jpeg. I could be tempted to activate the High ISO NR without Multi Shot NR set but because it would be permanently on I am a little concerned about a possible adverse impact on images taken at low ISO speeds.

As for Long Exposure NR, I need to remember the manual's suggestion that for most purposes the Auto setting will work well and will be performed automatically for any exposure longer than 1 second. I could be tempted to make that my default, too.

[Rufus]

What raw processor do you use and what are its noise reduction tools capable of doing ?

As an alternative I find selecting the black sky and bumping up the black level gets rid of noise better in situations like these.

I use the Lightroom RAW processor and often do additional work in Photoshop. I have to admit that on this image I did not use the NR tool.

Bumping up the black level would be a nice simple solution, so I like that idea.

[DanK]

I see the noise in the sky quite clearly on my larger monitor.

Some very informative answers, but I think a few key things are missing still.

would one expect the noise to be similar for any combination of aperture and shutter speed that gives a consant exposure value and a central meter reading?

No. What matters is exposure and amplification, not the settings you use to obtain a given exposure.

These two forms of noise reduction have entirely different effects. What Canon is now calling high ISO NR is the same genre as noise reduction in post-processing software: it hides random noise by smoothing out some details. It affects only JPEGs, as Bill pointed out. It inevitably loses detail.

Long exposure noise reduction (I'll abbreviate it as LENR; also called black-frame NR or subtractive NR) is designed to offset fixed-pattern noise that occurs when a sensor is on for a long time. It affects raw as well as JPEG images. It has no effect on random shot noise and therefore would have had no effect on your noisy sky. It works by taking a second exposure of identical length with the shutter closed and subtracting it from the first exposure. That eliminates hot pixels.

Your photo has incomplete exif, so I don't know the shutter speed, but from my experience taking night photos, I would guess that this is a fairly short exposure given that you were shooting at ISO 1600. With your camera, I suspect that LENR would have no effect at all; I doubt your exposure was anywhere nearly long enough to generate fixed-pattern noise. Your camera probably has an "auto" setting for LENR, and if so, I'm guessing that it would not trigger LENR for that shot. I do a fair amount of urban night photography at much lower ISOs, and hence longer shutter speeds, and I don't think the auto setting on my camera has triggered LENR a single time when I am doing that kind of photography. I routinely use it when I am taking night shots in settings that require exposures longer than a minute or so.

You mentioned not exposing longer than 1 second to avoid problems with the sensor overheating. Overheating depends on the circumstances (worse on a hot, humid summer day than on a cold autumn evening), but I would be surprised if you had any problems with overheating with exposures shorter than 10 minutes. Even when I was shooting with a 50D, a relatively noisy camera that was more prone to overheating than my current FF camera, I had to go over 10 minutes on a summer night to have problems. Here is a 50D shot that was an exposure of just under 10 minutes:



The real solution to the noise problem in photos like the one you posted is to shoot at a low ISO. Buildings, after all, don't move, so you have as much time as you want. When I shoot urban night photography, I most often shoot at ISO 100 and rarely go over 200. (I did find one I shot at 400, but I think that was simply a mistake.) If you also avoid exposing to the left, this should give you photos that need no noise reduction at all.

[Manfred M]

2. "Long Exposure Noise Reduction" does change the raw data.

This is my understanding as well Bill, but I'm not sure if it is universally correct. There was an outcry about Sony cameras a few years ago where people were complaining about a firmware update (and new model cameras) that "ate stars", i.e. small dots of light were likely captured but did not end up showing up in the raw data.

The second part of this is to remember that long exposure noise reduction uses a technique referred to as "dark frame subtraction", which is quite effective in identifying fixed hot pixels that are associated with long exposure image making. Most digital cameras allow for a maximum of a 30 second exposure. For longer exposures (done in "bulb mode") I use a manual workaround that emulates dark field subtraction.

The concept is actually quite simple; take an exposure the same length of time as the original shot but with the shutter closed. Any fixed pattern hot pixels will be recorded and the firmware in the camera will subtract the value of the hot pixels from the raw data before the raw file is written. The only downside is that it doubles the length of the exposure.

A simple workaround I use is to leave the long exposure noise reduction turned OFF and then to take a single shot (or series of shots) for the same exposure length that I have been using the the lens cap in place and the eye piece turned off on my DSLR. I take the original raw data and the dark field data and open both in Photoshop on separate layers. Using the Subtract blending mode, I remove the hot pixels from the working file.

The main advantage of the technique is that the exposure time for individual shots is not doubled like it when the in-camera long exposure noise reduction is used as well as I can use it for exposures longer than 30 seconds.

[DanK]

This is my understanding as well Bill, but I'm not sure if it is universally correct.

It's correct for Canon cameras. That is the only brand on which I have used LENR.

The main advantage of the technique is that the exposure time for individual shots is not doubled like it when the in-camera long exposure noise reduction is used as well as I can use it for exposures longer than 30 seconds.

I've never used a Nikon, but with Canon cameras, LENR is is not limited to 30 seconds. The black-frame exposure is determined by how long the shutter is open, even in bulb mode. I routinely use it for exposures longer than 30 seconds. The camera's computer remembers how long the shutter was open and does a black-frame exposure of the same length.

It is true, however, that this can be a nuisance, when one is taking exposures as long as 10 minutes each...

[Rufus]

Thank you Dan and Manfred for your recent posts. They don't call for a reply as such but they have certainly improved my understanding. Thanks again!

[proseak]

What does the histogram look like when you first open the file in Lightroom?

[Rufus]

What does the histogram look like when you first open the file in Lightroom?

Peter, here is the histogram when I open the original file in Lightroom:


There is much more noise evident in that original file than the processed one.

[Manfred M]

I've never used a Nikon, but with Canon cameras, LENR is is not limited to 30 seconds. The black-frame exposure is determined by how long the shutter is open, even in bulb mode. I routinely use it for exposures longer than 30 seconds.

Thanks Dan. I was wrong. Nikon works the same, although I'm not sure if that is something more recent or if the old cameras did so too or if there is a limit to how long an exposure can be handled. There is nothing in the manual to indicate what the maximum time is.

Regardless, I prefer my technique in doing a dark frame exposure manually and correcting in Post. That way I don't have to wait for minutes to do repeat exposures.

[William W]

Hi David,

it appears that I wasn’t clear in my previous responses to your questions. For clarity now here are two general points regarding my previous comments in this conversation:

1. All my responses were referring exclusively to CANON EOS Series DSLRs

2. As far as I know, once CANON creates, names and installs a function, within any Camera Series, ("EOS DSLR" is a "Camera Series") then that functionality is the same for every camera within that series, (and Canon tends to cut and paste the paragraphs in their User Instruction Manuals, too)

(If Manfred and Dan could also please note, as a comment to their Posts #14 and #15)

***

. . . Your camera may differ but mine definitely does not permit High ISO NR with Multi Shot NR when the file type is set for RAW+Jpeg. . . .

I use Canon EOS DSLRs, whilst I haven't used a 6DMkII, I am familiar with the use of the Camera Functions that we are discussing. I understood the underlined bit above - I simply didn't detail that understanding.

. . . I could be tempted to activate the High ISO NR without Multi Shot NR set but because it would be permanently on I am a little concerned about a possible adverse impact on images taken at low ISO speeds. . .

Again I think that my meaning was not clear: my point was that IF we select:

a) "High ISO NR without Multi Shot NR set"
and -
b) Capture to: "raw + JPEG (L)"
Then yes, sure the JPEG image is constantly and permanently affected by the NR selection, but if we are using a third party raw converter, then the raw file may be accessed without any effect of the NR Flag.

***

I think that the differentiation of the two types of Noise Reduction Canon offer and the explanations of those, which Dan gave in his Post #13, is excellent and goes to the core of the practical answer to your concerns.

WW