Lighting conditions and ISO noise!?

[Alfredo Fernández]

Hi,
i normally use my d300s at iso800 and normaly (sunny days, little overcaste) find the results pretty clean (at least for me).
yesterday i went to the beach to photograph some terns and stuff and it was very foggy.
when i cheked the images in my computer i found that the noise levels were unacceptable.
is there any relation between lightening conditions and noise levels?
thanks a lot,
Alfredo

[Mark Farnan]

Hi,
i normally use my d300s at iso800 and normaly (sunny days, little overcaste) find the results pretty clean (at least for me).
yesterday i went to the beach to photograph some terns and stuff and it was very foggy.
when i cheked the images in my computer i found that the noise levels were unacceptable.
is there any relation between lightening conditions and noise levels?
thanks a lot,
Alfredo

One answer is, sort of :). It isn't a direct relationship, but in my experience, is related to two things caused by lower, or softer, lighting conditions.

1. Where the information is happening in the Histogram

Check where the data is falling in the Histogram. On a brighter day (or more direct light), It is likley more of the information is higher up. I.e. the 'curve' peaks more on the right side of available data, than the left. Exactly WHERE it is, is of course exposure dependant, but I'm more refering to where the data may fall. There is always less noise in the higher stops, than the lower ones as a rule.

Noise is basically, varying information in the lower 'bits' of the sensor reading that is not the image itself. As Higher stops have more 'bits' per stop of information, the noise is less noticeable.


2. Overall Exposure.

Another factor of Noise is how long the sensor is ON, and Dimmer the light, longer the exposure is likley to be. I'm going to guess that on the foggy day, your overall exposure time was a lot longer. (Shutter speed basically). Even a couple of stops of shutter can make a big different at higher ISO's.

Note: The above is a simplification of a couple of the points, and is based on what i've seen shooting higher ISO's, but hope it helps.

Regards

Mark.

[William Malacarne]

Alfredo

See if this helps on exposure.

http://www.luminous-landscape.com/tutorials/expose-right.shtml

Bill

[Alfredo Fernández]

Thanks for the quick replies.
i always work on manual exposure and i know how/why to expose to the right, but thanks anyways.
ive been cheking the photos again, and can (heavy) mist cause REALLY soft focus ?? and are the OOF areas more vulnerable to noise?
thanks a lot,
A

[Roger Clark]

2. Overall Exposure.

Another factor of Noise is how long the sensor is ON, and Dimmer the light, longer the exposure is likley to be. I'm going to guess that on the foggy day, your overall exposure time was a lot longer. (Shutter speed basically). Even a couple of stops of shutter can make a big different at higher ISO's.

Note: The above is a simplification of a couple of the points, and is based on what i've seen shooting higher ISO's, but hope it helps.


Mark,
I agree with your first points, but not the above, at least for the length of time in any daylight scene. To get increased noise, one needs to go tens of seconds to many minutes exposure before thermal noise from the sensor would build up. Otherwise, noise is square root proportional to total exposure, regardless of the length of the exposure.

Alfredo,

Most of the noise we see in digital camera images is due to photon noise. Photon noise is the square root of the total photons captured. What this means is our perception of noise goes up with the square root of the exposure. By working at ISO 800 in a daytime scene, you have 8 times shorter exposure than if you metered at ISO 100. Thus you have square root 8 (=2.8) times worse signal-to-noise at all levels. Our perception of noise depends on the scene. On a foggy day, there is less contrast and with less contrast noise is more objectionable. For example, we often see noise in the sky in our images because the sky is low contrast with no fine details. If you stretch a foggy image to improve the contrast, you amplify the noise, making it quite apparent.

To minimize noise, use the lowest ISO that can give the shutter speed you need to stop an action you want stopped.

Roger

[WIlliam Maroldo]

Alfredo; did you read
http://www.luminous-landscape.com/tu...se-right.shtml?


I read it a while back, after I had already adopted the "expose to the right" method. What it doesn't mention is the relationship of contrast and the ability to "push" exposure.
If you compare the histograms of normally exposed images shot in high contrast light vs low contrast light, you'll notice that in high contrast light you'll see peaks that appear on most of the graph from right to left, and often end abruptly at one or the other end, or both. This is called clipping. The ability to shift the entire histogram to the right is often limited in high contrast light, and increasing exposure even moderate amounts can clip the highlights.
If you examine the histogram of low contrast light, which would be images shot under cloudy or overcast conditions, you'll notice that the histogram peaks do not fill the entire graph and that clipping isn't happening. Since the distance from darkest to lightest is so much shorter on the histogram, it allows the histogram to be shifted toward overexposure a great deal before clipping occurs.

The luminous landscape article does make important points about sensor behavior. The right most 1/5 of the histogram has 50% of the sensor's capability to capture digital information, and you want as much of the brightness range of the image to be in that area as possible, not only because of noise considerations, but because you will be ble to capture as much detail as possible.

Under the conditions you mentioned, foggy, you should be able to avoid digital noise a great deal, and it has nothing to due with parts being OOF, but because it is very low contrast light. Your images will definitely look washed out, and as you look at the camera LCD you will swear there is no detail there, but as long as you haven't clipped the highlights too much, you will be amazed how much is there when you drop exposure down to normal in the RAW converter, and even if you have clipped a few highlights using recovery sliders often works well. One thing to note: in post processing contrast needs to be increased, which can be done quite successfully. Also I have noticed that in low contrast light clipped highlight recovery works much better than in high contrast light.
If you exposed your foggy image normaly digital noise will certainly be a major problem with higher ISOs.
There is one thing that seems counter-intuative. In order to push exposure you need to either slow the shutter-speed, open up the aperture, or increase the ISO. The first two have limitations; you need a specific shutter-speed and aperture for the best shot, so increasing the ISO is usually the best choice.
You will actually need to increase the ISO to reduce noise. Sounds crazy, but it works.
This is because you will have less noise at lets say ISO 3200 when pushing exposure that at ISO 1000 using normal exposure.
Hope I've been helpful.~Bill

[Steven Kersting]

Thanks for the quick replies.
i always work on manual exposure and i know how/why to expose to the right, but thanks anyways.
ive been cheking the photos again, and can (heavy) mist cause REALLY soft focus ?? and are the OOF areas more vulnerable to noise?
thanks a lot,
A

Mist can definitely cause soft focus and a lack of contrast (moisture in the air causes light diffraction).

Noise is very susceptible to sharpening and exposure increases in post.

Noise is more apparent in areas without detail because it becomes the "detail". Detail in an image "hides" noise. Sharpening boosts the "detail" and therefore the noise. Selective sharpening is the only suitable compromise (and may not work).

Noise is also most apparent in darker areas which are amplified by higher ISO into a medium/high level. Areas which remain "black" will show the least noise. (This is opposite the long exposure noise behavior because the cause is "opposite") If most of the image is "flat", then adjusting the exposure to increase brightness in areas will increase the noise as well in those areas (until the point of clipping)....

[Roger Clark]

There is one thing that seems counter-intuative. In order to push exposure you need to either slow the shutter-speed, open up the aperture, or increase the ISO. The first two have limitations; you need a specific shutter-speed and aperture for the best shot, so increasing the ISO is usually the best choice.
You will actually need to increase the ISO to reduce noise. Sounds crazy, but it works.
This is because you will have less noise at lets say ISO 3200 when pushing exposure that at ISO 1000 using normal exposure.


Bill,
Most of the noise is proportional to the square root of the amount of light gathered. Thus, when one slows the shutter speed, or opens the aperture, more light is recorded. When one boosts the ISO, there is no change in light, and with normally metered exposures, we are exposing at a faster shutter speed so less light is recorded. With less light, the signal-to-noise ratio drops, and we perceive that as more noise in the image.

If one exposes at say 1/3200 second at ISO 3200 and you do the same 1/3200 second exposure at ISO 1000, the signal and photon noise are the same, thus the signal-to-noise ratio will be the same over the top 9 or so stops. In the deepest shadows a small additional noise will be apparent (due to read noise) at the ISO 1000, but it is an extremely small effect. If the ISO3200 image is made at any faster shutter speed faster than ISO1000 (for the same aperture), then the ISO 3200 will have worse perceived noise.

Roger

[Steven Kersting]

I think the confusion is that there is a difference between sensor noise and ISO noise.

Sensor noise (red/blue/hot pixels) is most apparent in dark areas and becomes a factor with long exposures (usually over 10s)

ISO noise is a different animal.

[Roger Clark]

I think the confusion is that there is a difference between sensor noise and ISO noise.

Sensor noise (red/blue/hot pixels) is most apparent in dark areas and becomes a factor with long exposures (usually over 10s)

ISO noise is a different animal.

Steven,
I'm not sure what you mean about "ISO noise." ISO doesn't have noise.

In all exposures there are potentially hot/dead pixels, some of which are mapped out so we don't see them. "Sensor noise" is read noise from the on-chip amplifiers. Then A/D converter noise gets added onto that signal. Both of those are at a very very low level, down 11.5+ stops for lower ISOs (below about 800). ISO is only gain, amplifying both signal and read+A/D noise. Higher ISO = higher gain.

Above a stop or two higher than complete black, the noise we see in digital camera images is dominated by photon noise--noise from the short interval of an exposure and the random arrival times of photons.

As one raises ISO, the read noise becomes a greater proportion of total signal (but still at a very low level), noise from the A/D proportionally decreases, and because fewer photons are typically collected, the signal-to-noise ratio in the signal (our images) decreases. But the dominant noise we perceive in digital camera images is not due to camera electronics, it is due to photons and Poisson counting statistics. That is a good thing, and is called photon noise limited. The solution to higher signal-to-noise images is then to collect more photons (longer exposure and/or larger lens aperture).

The other factor in long exposure images is thermal noise, which one can see in hot environments with a few seconds exposure (e.g. 90 degrees F and hotter). Colder environments require many minutes before thermal noise becomes much of a factor (e.g. near freezing).

Roger

[John Chardine]

Roger- thanks. I am really interested in this part of digital photography, the same way I need to know how my motorcycle engine works and how it's put together.

I think I understand the constraints placed on collecting photons when you have small sensor sites, and that we are almost at theoretical limits with current sensors. But as I understand it there is still a lot to be gained in the efficiency of sensors. Correct me if I'm wrong (I know you will but I had to say it!), but right now digital camera sensors throw away more photons than they collect (70-80%?). What is your understanding of how this pretty poor efficiency will improve in the future? I understand the Canon 1DIV is more efficient that most other commercial sensors out there so Canon have figured out how to do it to some degree. What are the future prospects? I assume if you doubled the efficiency of a sensor you would experience a lot less noise, AOTBE (all other things being equal).

[Steven Kersting]

Steven,
I'm not sure what you mean about "ISO noise." ISO doesn't have noise.
Roger

Any noise relevant to ISO ("photon noise"/amplifier noise) I've been in the habit of calling ISO noise.

Any noise due to long exposure (thermal noise) I've been in the habit of calling sensor noise.

I can see this is technically erroneous and oversimplification...but it works for me. :bg3:

[Roger Clark]

Roger- thanks. I am really interested in this part of digital photography, the same way I need to know how my motorcycle engine works and how it's put together.

I think I understand the constraints placed on collecting photons when you have small sensor sites, and that we are almost at theoretical limits with current sensors. But as I understand it there is still a lot to be gained in the efficiency of sensors. Correct me if I'm wrong (I know you will but I had to say it!), but right now digital camera sensors throw away more photons than they collect (70-80%?). What is your understanding of how this pretty poor efficiency will improve in the future? I understand the Canon 1DIV is more efficient that most other commercial sensors out there so Canon have figured out how to do it to some degree. What are the future prospects? I assume if you doubled the efficiency of a sensor you would experience a lot less noise, AOTBE (all other things being equal).

Hi John,
Taking the camera system as a whole, there are a lot of losses. The light first goes through the lens (perhaps 70% transmission; 30% loss), the the IR filter (perhaps 90% transmission), then the Bayer filters (peak at perhaps 80% transmission), then the blur filter (~95% transmission), then the transmission of the microlenses over the pixels (~95%), then the sensor is not 100% light sensitive, so some light hits non-sensitive areas (this is called the fill factor) and with the focusing the micro lenses, the equivalent fill factors are probably around 90%.

So the light hittin the sensor is down to 0.7 * 0.9 * 0.8 * 0.95 * 0.95 * 0.9 = 0.4.

Sensors are on the order of 40% efficient (40% quantum efficiency, QE), so of the light incident on the lens, about 0.4 * 0.4 = 0.16 or 16% (round off to 20%) is actually detected.

Quantum efficiency could be improved to around 90 to 95% with current technology (thinned, back-side illuminated sensors). Large sensors can't be mass produced easily, but manufacturing may only add a thousand or two dollars (amateur astronomers are using thinned back-side illuminated CCDs for example). I'm actually surprised Canon, Nikon, or Sony have not come out with a super pro model with increased QE. Imagine a 1D Mark IVq with double the quantum efficiency. What one can do now at ISO 400 could be done at ISO 800 with the same noise.

Trying to improve the other components in the system is tough.

A new technology sensor that doesn't need micro-lenses, or RGB or IR filters and each photon's energy could be individually measured. Then one could synthesize the color profile that matched the eye and improve the system efficiency by several fold. There is nothing I know of in labs that can do this yet, so probably not something photographers will see anytime soon.

But digital is way ahead of film. QE of film is only 1 to 2%, so we should be happy (I am):S3::S3::S3:

Roger

[Tom Graham]

Most interesting Roger, thanks. I had no idea that film QE was 1-2%, makes our digital sensor at (low?) 20% look great. Just for fun, from 2% to 20% would be what, 10 times better, or square root of 10 f stops better, or about 3 f stops?
Tom

[John Chardine]

Thanks Roger. So I understand the calculations:

1. The 40% QE comes from the "light hitting the sensor = 0.4" calculation above?
2. Why do you square the QE to get sensor efficiency (0.4 * 0.4 = 0.16)?

[Roger Clark]

Thanks Roger. So I understand the calculations:

1. The 40% QE comes from the "light hitting the sensor = 0.4" calculation above?


Hi John,

No, the two numbers just coincidently came out the same. The light going through the optical
system, including the microlens just above the photo-sensitive pixel get 60% absorbed, so system throughput is 40%. Say you have 1000 photons incident on the front of the lens that will be focused on a pixel. Only about 400 photons make it to the silicon.



2. Why do you square the QE to get sensor efficiency (0.4 * 0.4 = 0.16)?

Continuing the example, 400 photons reach the silicon pixel. Of those, about 40% actually are absorbed and generate an electron that can be measured. That is the quantum efficiency. So of the original 1000 photons, about 40% make ti to the pixel, and the pixel only captures about 40% of those photons, so .4*.4 = .16 or 16%.

Does that help?

Roger

[Roger Clark]

Most interesting Roger, thanks. I had no idea that film QE was 1-2%, makes our digital sensor at (low?) 20% look great. Just for fun, from 2% to 20% would be what, 10 times better, or square root of 10 f stops better, or about 3 f stops?
Tom

Tom,
That is correct. The increased QE of digital sensors is the main reason we can image at much higher ISOs.

Roger

[John Chardine]

Makes sense. It was the two 0.4s that were confusing me. Thanks Roger.