Confusion over high ISO, noise, and light levels

Messages
11,513
Name
Stewart
Edit My Images
Yes
I'm confused. I'd be grateful if someone could help me become unconfused.


The background

There's a discussion on another thread (which I don't want to hijack) which has touched on the suitability or otherwise of a Nikon D90 for low-light photography such as weddings.

One contributor argued that ISO 3200 is quite useable:
This is at 3200 on the kit lens (18-105) straight out the camera with no noise reduction

sample3200.jpg



Not perfect but usable

That photo was actually taken in decent light, and using ISO 3200 was an accident by the photographer. But then lots of other people weighed in to say that it in low light there would be much more noise:
... with the light you have noise will be less than if taken in a dark church.
... in a brighter room the noise will be far less than in a dark room.
Yes totally agree
I see this quite a lot, even some of the bloggers who test new kit show ISO tests in a lit room in normal circumstances and say 'Yeah no noise at ISO6400 blah blah this camera is really great blah blah"

It all changes when you turn the lights off....
Correct, there is a difference in taking a shot at 3200ISO in nice light or a dark church.

Why I'm confused

The photo which I've included above was taken at {ISO 3200, f/8, 1/90th}. I'd say it looks quite reasonably exposed, so those parameters are clearly about right for the prevailing light level.

Now, suppose that shot had been taken in a dim church where there was 5 stops less light (ie only about 3% of the light available in that photo). But suppose that the photographer also had an f/1.4 lens available to him. By a fortunate coincidence f/1.4 is 5 stops faster than the f/8 he actually used. So he could shoot at {ISO 3200, f/1.4, 1/90th} and the exposure would be correct.

So ... why is {ISO 3200, f/1.4, 1/90th} noisier than {ISO 3200, f/8, 1/90th}, if the two photos are both exposed correctly?
 
Noise is more visible in shadows. So in a dark church there is much more shadow and dark areas.

Getting a well exposed image will reduce the noise though - but it will be a lot more noisy.

PS you should have posted this on the same thread - not fair linking someone elses image like this. - Change it to a link rather than use the img.
 
Last edited:
Noise is more visible in shadows. So in a dark church there is much more shadow and dark areas.

Getting a well exposed image will reduce the noise though - but it will be a lot more noisy.

PS you should have posted this on the same thread - not fair linking someone elses image like this. - Change it to a link rather than use the img.

Pretty much what I would say, a nailed exposure will help reduce noise but only to a certain extent, depending on the sensors capabilities. No mater how good an exsposure is noise is always more noticable in areas of shadow or dark colours.
 
Noise is more visible in shadows.
Yes, I understand this, and I know why.
So in a dark church there is much more shadow and dark areas.
But this is where you've lost me. I don't think you've addressed my question, which is this:

If you take one photo in decent light at {ISO 3200, f/1.4, 1/90th}, and one in poor light at {ISO 3200, f/8, 1/90th}, and both are exposed correctly, why does the latter exhibit more noise?
 
Look at the colour of the background and the colour of the light

In a dimly lit church it's well..... just dark - Yes you can get a well exposed skin tone but the rest of the image will be noisy because it's darker.

I suppose it depends on the situation.
 
Look at the colour of the background and the colour of the light

In a dimly lit church it's well..... just dark - Yes you can get a well exposed skin tone but the rest of the image will be noisy because it's darker.

I suppose it depends on the situation.
So if | understand you correctly, what you're saying is that it's the DIFFERENCE between the light tones and dark tones that's important, but the actual quantity of light isn't. Right?
 
In brief, it is to do with how digital sensors record different tonal values Stewart.

Lets say the sensor has a six stop dynamic range, recording a 12bit image, recording 4096 tonal values. The assumption would be each f stop of the six stop range would record equal amounts of tonal value, this is not so, each stop records half the light of the previous one.

Half the levels are devoted to the brightest stop (2048), half of the remainder (1024) to the next stop, and so on until the last and darkest area only has 64 levels.

In trying to boost shadows you only have 64 levels in the darkest areas, as there are fewer tonal values to spread the information over, the changes are more abrupt and obvious.

In the brighter areas there are 2048 tonal values available so any change is less abrupt, so hardly noticed.

This led to wise photographers exposing "to the right' to capture more detail in the available 2048. The trick is to err towards over exposure without clipping the highlights.
 
Not the difference. Basically it is harder to see noise on a light background, such as the photo above. If the image has many darker areas, such as a dim church or gig setting, there will be a lot more darker colours and black areas. These dark areas are where you are able to see the noise.

A simple way to illustrate is take a picture at max iso of a black piece of paper and a white piece of paper. You will see the noise a lot clearer on the black sheet due to how noise shows itself.

It's not the difference between light and dark tones that make the difference, it is the amount of dark tones where the noise artefacts show up clearer
 
In brief, it is to do with how digital sensors record different tonal values Stewart....
Thanks Martyn, but I'm afraid that doesn't have anything to do with my question.

You've explained - very well - why there is more noise in (relatively) dark areas than in (relatively) light areas.

But that doesn't say anything about whether there is more noise in low light than in decent light when the photograph is exposed correctly.
 
In brief, it is to do with how digital sensors record different tonal values Stewart.

Lets say the sensor has a six stop dynamic range, recording a 12bit image, recording 4096 tonal values. The assumption would be each f stop of the six stop range would record equal amounts of tonal value, this is not so, each stop records half the light of the previous one.

Half the levels are devoted to the brightest stop (2048), half of the remainder (1024) to the next stop, and so on until the last and darkest area only has 64 levels.

In trying to boost shadows you only have 64 levels in the darkest areas, as there are fewer tonal values to spread the information over, the changes are more abrupt and obvious.

In the brighter areas there are 2048 tonal values available so any change is less abrupt, so hardly noticed.

This led to wise photographers exposing "to the right' to capture more detail in the available 2048. The trick is to err towards over exposure without clipping the highlights.

A much better explanation than mine :thumbs:
 
A simple way to illustrate is take a picture at max iso of a black piece of paper and a white piece of paper. You will see the noise a lot clearer on the black sheet due to how noise shows itself.
Interesting thought experiment, but I don't think it's relevant.

What everyone was saying in my original post is that if you take a picture at max ISO of a black piece of paper in a well lit room, and a black piece of paper in a dimly lit room, then there's a difference.

Is there? And if so why?
 
I suggest you try shooting in your living room during the day and then do the same at night both with a correct exposure and both at the high ISO level - You will see the difference.
 
Thanks Martyn, but I'm afraid that doesn't have anything to do with my question.

You've explained - very well - why there is more noise in (relatively) dark areas than in (relatively) light areas.

But that doesn't say anything about whether there is more noise in low light than in decent light when the photograph is exposed correctly.

The relatively dark areas only have 64 tonal values, the bright correctly exposed area will have 2048 tonal values.

In the image above there is a far greater area using the 2048 values, than if a similar image had been shot in a dark church, which would have a far greater area only using 64 tonal values.
 
I suggest you try shooting in your living room during the day and then do the same at night both with a correct exposure and both at the high ISO level - You will see the difference.
I'm not arguing that there isn't a difference, Jim. So many people who are more experienced and more capable photographers than me have said there is a difference.

It's just that I can't understand WHY there should be a difference. (And therefore, since I'm a logical person, that does make me start to doubt whether there really is a difference.)

So far everybody who has contributed to this thread has either:
(a) asserted that there is a difference, without explaining why; and/or
(b) explained why there is a difference between noise levels in light areas and shadows, which is a totally different issue.

I think I might have to try the experiment over the weekend, unless someone can come up with an explanation.
 
... if a similar image had been shot in a dark church, [it] would have a far greater area only using 64 tonal values.
Why, if the image is exposed identically to the image above? (i.e. With different but appropriate settings, so that the dark areas are just as dark as in the image above, and the light areas are just as light as in the image above.)
 
Last edited:
Interesting thought experiment, but I don't think it's relevant.

What everyone was saying in my original post is that if you take a picture at max ISO of a black piece of paper in a well lit room, and a black piece of paper in a dimly lit room, then there's a difference.

Is there? And if so why?

I might have this a little wrong but how I see it.......

Generally speaking, you'd only use a high ISO if there isn't enough light to expose the shot correctly for the shutter/f settings you want to use but in the example of taking a shot of a something completely black, regardless of shutter speed, f-stop and even if the exposure was spot on, the higher ISO setting would give more noise in the shot.

The more technical explaination above I kinda get (and don't dispute). Thinking of it another way..... If you had an old audio recording on a tape and wanted to copy it to a new tape and make it a little louder, to make it louder you would also increase the amount of background noise on the new recording.
 
Generally speaking, you'd only use a high ISO if there isn't enough light to expose the shot correctly for the shutter/f settings you want to use but in the example of taking a shot of a something completely black, regardless of shutter speed, f-stop and even if the exposure was spot on, the higher ISO setting would give more noise in the shot.

The more technical explaination above I kinda get (and don't dispute). Thinking of it another way..... If you had an old audio recording on a tape and wanted to copy it to a new tape and make it a little louder, to make it louder you would also increase the amount of background noise on the new recording.
What you're saying is that higher ISO settings give more noise. But that isn't disputed.
 
Why, if the image is exposed identically to the image above?

If the image is exposed, and importantly lit identically to the image above it would look identical. The point raised in the thread you took the image from made the valid point that an image for a wedding, taken in a church would not be lit identically, it would have a far higher proportion of dark areas.
 
What you're saying is that higher ISO settings give more noise. But that isn't disputed.

But in my example of an audio recording.......

Where there's a lot of volume in the actual music (the "light bits") the "hiss" or noise isn't as noticable as the hiss during breaks in the music (the "dark bits")
 
So ... why is {ISO 3200, f/1.4, 1/90th} noisier than {ISO 3200, f/8, 1/90th}, if the two photos are both exposed correctly?



In the f1.4 photo your scene is darker.
.. but those rays of reflected light rays are both spread and narrow rays rather than the shot at f8 where the rays are narrow and nearer parallel.

Im thinking in both scenarios your asking the sensor to collect identical amount of light, but at f8 the light is well defined throughout. At f1.4 the light has some rays in focus and some outside of that making the quantity overall less defined.

That’s my take on what might be happening ...simply down to the sensor recording less defined shadow detail because of the wide aperture.
 
If the image is exposed, and importantly lit identically to the image above it would look identical. The point raised in the thread you took the image from made the valid point that an image for a wedding, taken in a church would not be lit identically, it would have a far higher proportion of dark areas.
I think we're agreeing, Martyn. I think you're saying that it's the QUALITY of the light that matters (and specifically the arrangement of the light sources and the quantity of shadow they provide), rather than the QUANTITY of light.
 
In the f1.4 photo your scene is darker.
.. but those rays of reflected light rays are both spread and narrow rays rather than the shot at f8 where the rays are narrow and nearer parallel.

Im thinking in both scenarios your asking the sensor to collect identical amount of light, but at f8 the light is well defined throughout. At f1.4 the light has some rays in focus and some outside of that making the quantity overall less defined.
Thanks. That's a good attempt at an explanation and exactly the kind of thing I was after.

Now all I have to do is try and understand it ... and then think about whether I believe it ... but at the moment it's making my head hurt.
 
A few points....

1. Smooth or OOF areas show noise more than sharp and highly textured ones because there is nothing other than noise to be seen in the otherwise smooth tones. If you open up 5 stops from f/8 to f/1.4 then your reduced DOF will generally send far more of the scene OOF. That may make noise apparent where previously it was obscured by (or mistaken for) texture.

2. Compared to a typical strong source of light such as sunlight or flash, dim ambient light is normally unbalanced with respect to the spectral distribution. Tungten light starves the blue channel of data (photons) and that needs to be remedied by applying a white balance adjustment. Even if your balanced exposure looks correct you would have started with an underexposed blue channel at the moment of raw capture and the white balance correction would have had to increase the blue channel amplitude, including the noise. The same is true at the opposite end of the spectrum when you are shooting under heavily blue biased light such as open shade.

If you want to make a fair comparison of noise differences at 3200 ISO when adjusting your exposure by 5 stops (or any other figure) to compensate for reduced light levels, make sure the only thing about the light that is changing is the quantity, not the quality, not the colour. Probably the best way available to the layman to put this to the test is to use a quality flash gun or strobe as your sole source of light and fire off shots with the flash at full power and at 1/32. There may be a small shift in colour balance but at least you will be controlling the variables as far as possible. If you want to eliminate DOF effects too then simply choose a flat target angled parallel to the sensor plane.

EDIT : Maybe using 5 stops (or whatever you like) of ND filtering would be equally effective in helping eliminate unwanted variables. Then you could experiment with adjustments through aperture and/or shutter speed.
 
Last edited:
affiliate links in your sig need to go please Tim...


Stewart, its Gain...higher gain = more grain...
 
1. Smooth or OOF areas show noise more than sharp and highly textured ones because there is nothing other than noise to be seen in the otherwise smooth tones. If you open up 5 stops from f/8 to f/1.4 then your reduced DOF will generally send far more of the scene OOF. That may make noise apparent where previously it was obscured by (or mistaken for) texture.
That's brilliant, Tim. Very well explained, and quite believable. It's not immediately clear whether or not the effect is sufficiently large to account for the real-life observations which other people have made, but that can be tested. And your hypothesis is testable in other ways too, because it implies that:
  • if the difference in exposure is achieved by changing the shutter speed rather than the aperture, then this effect will not be present
  • if the photo is entirely out of focus, then this effect will not be present
I can see that my weekend experiments are becoming more complex....
2. Compared to a typical strong source of light such as sunlight or flash, dim ambient light is normally unbalanced with respect to the spectral distribution. Tungten light starves the blue channel of data (photons) and that needs to be remedied by applying a white balance adjustment. Even if your balanced exposure looks correct you would have started with an underexposed blue channel at the moment of raw capture and the white balance correction would have had to increase the blue channel amplitude, including the noise. The same is true at the opposite end of the spectrum when you are shooting under heavily blue biased light such as open shade.
Also quite believable. Again it's a testable hypothesis:
  • if the nature of the light source is the same (ie it is on the quantity of light which varies, not the quality) then this effect will not be present.
Maybe using 5 stops (or whatever you like) of ND filtering would be equally effective in helping eliminate unwanted variables. Then you could experiment with adjustments through aperture and/or shutter speed.
Absolutely!
 
Now all I have to do is try and understand it ... and then think about whether I believe it ... but at the moment it's making my head hurt.

Thanks for making sence of it.

Its ok, no one likes my 'apature is a wide ray filter' angle.

A few points....

1. Smooth or OOF areas show noise more than sharp and highly textured ones because there is nothing other than noise to be seen in the otherwise smooth tones. If you open up 5 stops from f/8 to f/1.4 then your reduced DOF will generally send far more of the scene OOF. That may make noise apparent where previously it was obscured by (or mistaken for) texture.

2. Compared to a typical strong source of light such as sunlight or flash, dim ambient light is normally unbalanced with respect to the spectral distribution. Tungten light starves the blue channel of data (photons) and that needs to be remedied by applying a white balance adjustment. Even if your balanced exposure looks correct you would have started with an underexposed blue channel at the moment of raw capture and the white balance correction would have had to increase the blue channel amplitude, including the noise. The same is true at the opposite end of the spectrum when you are shooting under heavily blue biased light such as open shade.

Very interesting second point Tim ...how much percentage wise would you guess one against the other for making noise? ...just to get an idea of the scale of the problem.
 
Last edited:
Stewart, I look forward to seeing the results. I hope you will share. :)
 
That's brilliant, Tim. Very well explained, and quite believable. It's not immediately clear whether or not the effect is sufficiently large to account for the real-life observations which other people have made, but that can be tested. And your hypothesis is testable in other ways too, because it implies that:


Tims first explanation is exactly the same as mine btw. ... just so you know. :)
 
This is another sample at ISO 3200 and a much darker scene, this photo is bad and I was about to delete it when I thought it may be of use on this thread. It was taken in candlelight - one of the shots that did not work out

It was taken at F2 - 1/45 - ISO 3200

sampleISO2.jpg
 
Very interesting second point Tim ...how much percentage wise would you guess one against the other for making noise? ...just to get an idea of the scale of the problem.

I wouldn't have a clue how to quantify the effect in terms of perceived noise, but you can pull out the WB multipliers from most raw files for different settings. These will show that R and B channels are almost always weak compared to G and need some degree of boosting. Just how much depends on how far off balance the light source is.

Here is the table of WB adjustment factors from a file from my 7D....

20110121_154511_.JPG


Since in a Bayer sensor you actually get RGGB for each pixel you will see that the GG channels are always with a fixed multiplier value of 1024.

If you read down to the values for Daylight WB you see 2162 1024 1024 1440. That means that the red channel needs boosting by a factor of 2162/1024 = 2.11X with respect to green. Blue also needs boosting, but by a more modest 1440/1024 = 1.4X.

If we look at the values for a Tungsten adjustment we see 1522 1024 1024 2162. Now the red channel needs boosting by a more modest 1522/1024 = 1.49X but the blue now needs a boost of 2162/1024 = 2.11X. That's as bad as the red channel in daylight.

If we look at the values under shaded lighting we get 2515 1024 1024 1222. Now the red channel needs a boost of 2515/1024 = 2.46X and blue needs 1222/1024 = 1.19.

So switching between tungsten lighting and shaded conditions there is a big swing between the multipliers needed for the R and B channels. If you compound those natural (almost unavoidable) underexposures at the sensor with additional underexposure through poor camera operation then the potential for excessive noise is increased further.

I say "almost unavoidable" because there is a way to equalise the multipliers by using a magenta filter to reduce the sensitivity of the green channel. This allows you to increase your exposure (without blowing the green channel) and more fully saturate the red and blue channels, which will then not require further boosting. You do, however, then need a very special custom white balance in order to get an image that looks right. This custom white balance is know ans "Uni White Balance" and has multipliers of 1024 1024 1024 1024, equivalent to 1X for both red and blue relative to green.
 
Last edited:
Yes when looking at an image most of the noise is held in the blue channel I forgot about that!
 
That would be a tungsten lit image, I presume.

Any image will have more noise in the blue channel - Just at higher ISO it gets more pronounced.
 
Any image will have more noise in the blue channel - Just at higher ISO it gets more pronounced.

Really? Even in a light source of 10,000K?

I think you'll find the red channel is equally vulnerable to noise if the light source is short on red wavelengths. The red channel is the cause of noise in what should be clear, smooth, clear blue skies on a sunny day.
 
Why, if the image is exposed identically to the image above? (i.e. With different but appropriate settings, so that the dark areas are just as dark as in the image above, and the light areas are just as light as in the image above.)

But hang on - you're comparing a subject in a bright room, well lit, with possibly a bride (who's bright white) in a dimly lit church. Doesn't matter how you compare they are two different images. You'll expose for the bride at one end of the scale, and possibly leave larrge dar areas because.
 
It seems to me that what Stewart is saying is this:

  • Subject = A
  • Shot 1 of A is in good light and settings are ISO 3200, f8, 1/90s
  • Shot 2 of A is in dim light and settings are ISO 3200, f1.4, 1/90s
A higher ISO in either shot would increase noise.
A slower shutter speed in either shot would also increase noise.

However, shutter speed and ISO remain the same for both shots - only the aperture has changed. Therefore noise levels should remain the same in both shots.
 
Really? Even in a light source of 10,000K?

I think you'll find the red channel is equally vulnerable to noise if the light source is short on red wavelengths. The red channel is the cause of noise in what should be clear, smooth, clear blue skies on a sunny day.

I could easily be wrong :)
 
It seems to me that what Stewart is saying is this:

  • Subject = A
  • Shot 1 of A is in good light and settings are ISO 3200, f8, 1/90s
  • Shot 2 of A is in dim light and settings are ISO 3200, f1.4, 1/90s
A higher ISO in either shot would increase noise.
A slower shutter speed in either shot would also increase noise.

However, shutter speed and ISO remain the same for both shots - only the aperture has changed. Therefore noise levels should remain the same in both shots.


Possibly right - but noise will be far more visible in the dimly lit shot.
 
Back
Top