Photographic noise explained

I need to resurrect this old thread because after reading through the Topaz article, there are a few statements in there that are at best misleading leading to very common misconceptions, especially by newbies to photography, about what exposure actually is and what causes noise.

1. What Are The Sources Of Noise?

There are 2 main sources of noise in an image.

a) Shot Noise - which is the most common noise we see in a photo and is due to the random nature in which light arrives at the camera sensor.

b) Read Noise - noise introduced by the internal electronics and heat from the camera.

Shot noise is by far the largest component of the total noise in a photo.

2. What is "Exposure"?

Exposure is defined, in laymans' terms, as the amount of light that falls on the sensor per unit area.  Why per unit area? To take into account that cameras have different sensor sizes. 

Only scene lighting, aperture and shutter speed affect exposure. ISO by itself has no affect on exposure.

Exposure and image lightness are two totally different things and it's when someone sees the two as being the same thing that they are more likely to get themselves in a muddle and wonder why there is so much noise in their photos.

Maybe think of it this way.  For a given scene and scene lighting -

a) f/8, 1/500s, ISO 100 gives you an image that appears a little too dark for your liking so you bump up the ISO.

b) f/8, 1/500s, ISO 200 now gives you an image that looks how you like it without any highlight clipping

Settings a) and b) will produce exactly the same exposure because the same amount of light has hit the sensor per unit area but setting b) will output a lighter looking image.

In this situation ISO has had no effect at all on the exposure.  It just produced a lighter image because it applied an amplification to the output of each pixel on the camera sensor. 

Both Settings a) and b) will have the same amount of shot noise but b) will most likely have less read noise (see below for explanation).

Digital cameras are designed to output an image with an "average" 18% grey lightness.  In any of the semi-auto modes (Aperture and Shutter priorities), setting a higher ISO is telling the camera to then set a lower exposure using whichever exposure control it has been told to set - aperture or shutter speed.

You need to be aware that the larger the exposure, the larger will be the Signal to Noise Ratio (SNR).

3. What Causes Visible Noise In A Photo?

That answer is fairly simple.  It is a low SNR resulting from a low exposure that makes noise visible in an image.  You also need to distinguish between actual noise present in an image and the amount of visible noise in an image.

Think of it this way - a photo taken on a bright sunny day on a beach will have very much more noise actually in the photo than a photo taken indoors in very low lighting.  The reason being because the exposure (amount of light falling on the sensor per unit area) of the beach photo will be very much larger than the exposure of the low light indoor photo.

BUT, because the SNR of the beach photo is also very much larger than the SNR of the indoor photo, the noise in the beach photo will hardly, if at all, be visible.  Whereas the indoor photo with the much lower SNR will have much more visible noise even though it has less actual noise in the image than the beach photo.

4. A Common Misconception

"Raising ISO increases the sensor's sensitivity to light" is a common misconception and is simply not true.  

The camera's sensor is not some kind of vacuum cleaner that "sucks" in more light when you raise ISO as show in section 2 above.

The misconception of sensors having various sensitivities to light because of ISO stems from the film days when films did have varying sensitivities to light.  But digital sensors do not behave like film in this regard.

Raising ISO in many cameras actually reduces the read noise (as described earlier) according to the charts at photonstophotos.

5.  How to Minimise Visible Noise?

Given that it is a low SNR resulting from a low exposure that makes noise visible in an image then we need to maximise the exposure within our creative constraints without clipping important highlights to minimise the visible noise.  How you do that is totally up to you depending on your preferred photography styles and camera modes.

When hand-holding the camera, the easiest way I know how is to use Manual mode + Auto ISO.  I normally like to capture the Depth Of Field I want with no motion blur in the photo.

So I typically set the widest aperture (smallest f-number) that will give the DOF I want and the slowest shutter speed I can hand-hold at for the focal length I am using and freezes any movement without clipping important highlights.  I let the camera set the output image lightness via Auto ISO to whatever it thinks is appropriate.  I can always adjust the image lightness in post to what I like but obviously cannot adjust the exposure in post.

This way I have maximised the exposure (within my creative requirements of DOF and motion blur), maximised the SNR resulting in minimising the visible noise in the raw data file.

I hope this post will help someone improve their image quality regarding minimising visible noise in their photos, especially in low light situations.

Hi BuckSkin.   My comments are in red.

Hi Casper,
Thank you for a very lucid and enlightening article. SNR means so much more to me now.
I have always been told that ISO is like gain in electonics, an amplification of the signal.
This means that both the signal and the noise are amplified.
Not that I known any more about the physics of sound than I do about light.

Thank you Fauxtoto and yes, your "speculation" is spot on.

The concept behind ETTR (Exposure To The Right) is that by maximising the amount of light that hit the sensor within your creative restrictions, as I described in my earlier post, you maximise the SNR which results in minimising the visible noise in the image.

You gain the most benefit of ETTR when shooting raw, especially in but not restricted to low light situations, because of the large amount of extra data in a raw file compared to a jpeg.  

For anyone not sure what ETTR is, this is a summary of the ETTR technique.

1. Set the camera on a tripod and in LiveView with the RGB histogram displayed on the LCD screen.

2. Set the widest aperture (smallest f-number) that will give the DOF or whatever other creative intent you have.

3. Set ISO to base - the lowest natural ISO setting on your camera, not any of the extended low or high values your camera might have.

4. Assuming there is no motion in the scene, slow the shutter speed until the histogram data begins to touch the right side of the histogram indicating highlight clipping is about to occur.

5. Press the shutter release to take the photo.

This way you have maximised the amount of light that hit the sensor within your creative intent (step 2) without clipping highlights.  By maximising the amount of light to hit the sensor you have maximised the SNR and so will minimise visible noise in the final image.

Now, because you have slowed the shutter speed until highlight clipping is about to occur,  when you open the raw file in you raw converter (ACR or whatever) it will most likely look way to light/bright.

This is where you now turn down the image lightness, using the "exposure" slider or equivalent, to what looks nice to you.  But because your exposure (amount of light that hits the sensor per unit area) was much larger than if you simply had zeroed/centred the camera's meter needle when you took the shot (which would have resulted in a much lower exposure than using ETTR) there will be much less visible noise in the photo taken by ETTR.

For example -

For a given scene lighting -

Say you set ISO 100, f/8  and the camera sets 1/400s shutter speed in aperture priority to centre the meter needle and so give you a nice looking image.

Now, after taking the shot above you see the histogram shows there is still significant amount of highlight headroom - empty space between the right most histogram data and the right edge of the histogram.

Let's say you experiment with shutter speed and see that with ISO and aperture remaining the same that 1/200s pushes the histogram data to just touch the right edge of the histogram.  You have now added 1 stop of light onto the sensor and still without clipping highlights.

Now obviously the raw data with ISO 100, f/8 and 1/200s will look much lighter in the raw converter than the ISO 100, f/8 and 1/400s raw data that gave the nice looking image with the centred meter needle.  In the raw converter you then turn down the ISO 100, f/8 and 1/200s image lightness by 1 stop to give the image lightness you like.

But because the ISO 100, f/8 and 1/200s image had twice as much light hitting the sensor than the ISO 100, f/8 and 1/400s image, the ISO 100, f/8 and 1/200s image will have less visible noise in it than the ISO 100, f/8 and 1/400s image because of the larger exposure, even though both images now have the same image lightness in the raw converter.

I hope this makes sense and helps demonstrate the benefits of maximising the exposure within creative requirements without blowing highlights and hence the benefits of ETTR.





 

Hi popcorn,

That is very interesting because birds in flight is probably my least preferred genre but I dabble sometimes for the challenge.  I'm more into landscapes and events and chasing grandkids around all of which are much slower moving targets than BIF  .  So I am not an expert on BIF by any means but I can try to help.

You mention you shoot wide open with the birds mainly backlit, which to me explains why your skies are always blown out as you describe. Unless you have a specific reason to have the side of the bird facing you mainly in shadow perhaps try to position yourself so that the sun is mainly behind you so that the birds are lit on the side facing you.

These are two examples of BIF photos I have taken with my 150mm-650mm Tamron lens and a summary of what I do.

I set the camera in manual + Auto ISO with Evaluative/Matrix metering.

For BIF I normally set the aperture at which my lens is sharpest, normally ~2 stops down from wide open. With the Tamron lens I set f/9 in both photos which is about 2 stops down from wide open at f/5. DOF is not normally an issue for BIF at f/9.  I know from experience that for BIF I need something around 1/2000s - 1/3000s to hand-hold the lens at the longer focal lengths and freeze the motion (flapping wings etc) of the birds.  Your required shutter speeds might vary.

I then take a few test shots without any birds of the more likely backgrounds (sky, water, foliage etc) I am likely to encounter while photographing the birds and check that there is no clipping of highlights.  If the camera set base ISO and I still had highlight clipping then I would either close the aperture a bit more or more likely select a faster shutter speed until I have no clipping on the camera's histogram and/or blinkies/zebras.

Then I am good to go 

Especially with BIF where the backgrounds and scene lighting can vary significantly and the birds are not going to wait while I fiddle with camera settings I don't worry to much about nailing the maximum exposure or ETTR especially if the lighting is very good, a bright sunny day.  Maximising the exposure to minimise visible noise is more important in low light situations for the reasons I posted in an earlier post.

Anyway, I hope this helps since I am not really a BIF photographer.

40 to 50 years ago, the sportsman's catalogs and magazines were full of camera/rifle stocks of many makes and designs; several were even set up such that pulling the trigger snapped the shutter.

I got to thinking about them a couple years ago and I think I found one source for a store-bought version.

All of those wonderful magical catalog merchants like Herters are now gone the way of most good things

The gunstock idea and principle is great; I don't know why they aren't as popular now as then.

I dearly love the feel of a rifle-stock tight against my shoulder.

When I first started out I used to set ISO first according to the ambient lighting, then aperture or shutter speed.

Basically on a bright sunny day I would set ISO 100, Cloudy/Overcast 200 or 400, Indoors 400 - 1600, Very Low Light 3200+ and use Aperture priority.

I now know that is not necessarily the best way to minimise visible noise in an image.  The danger is that if I choose a higher ISO than I really need to get the output image lightness I want then the camera will have set a lower exposure, due to the higher ISO, by setting a faster shutter speed to output an average 18% grey that cameras are calibrated to.  And as I posted earlier, it is the low SNR resulting from a low exposure that makes noise visible, not ISO by itself.

Many newbies come to the conclusion that setting a higher ISO creates noise whereas if they accept the definition of exposure as being the amount of light that hits the sensor per unit area then obviously ISO by itself cannot create noise because the effect of ISO is applied after the shutter has closed and no more light can hit the sensor - ref. Point 2 in my first post.

Much of the "teaching" about the Exposure Triangle, at least to my understanding, says that for a given exposure if you double one of ISO, aperture or shutter speed then you need to halve one of the other 2.  That again is not correct if you accept the definition of exposure as being the amount of light that hits the sensor per unit area but unfortunately is a very common misconception.

So, my advice is to abandon the concept of the Exposure Triangle.  

Remember, as I posted earlier, exposure and image lightness are two totally different things and it is when people see them as the same that they can start scratching their heads wondering why they see so much noise in their photos.

The way I look at it is that for a given image lightness, not exposure, there is a reciprocal relationship between any two of aperture, shutter speed and ISO.  I don't need a triangle or any other shape to understand that.

For example if you go from ISO 200, f/8, 1/200s to ISO 400, f/8, 1/400s then the output image lightness will be the same but the exposure will have been halved in the 1/400s image resulting in a lower SNR causing more visible noise in the 1/400s image even though both images will have the same lightness.

Hopefully this is at least some food for thought if not actually clearing up a common misconception.