Noise and the Camera Sensor

Camera sensors all suffer with two major afflictions; diffraction and noise; and between them these two afflictions cause more consternation amongst photographers than anything else.

In this post I’m going to concentrate on NOISE, that most feared of sensor afflictions, and its biggest influencer – LIGHT, and its properties.

What Is Light?

As humans we perceive light as being a constant continuous stream or flow of electromagnetic energy, but it isn’t!   Instead of flowing like water it behaves more like rain, or indeed, bullets from a machine gun!   Here’s a very basic physics lesson:

Below is a diagram showing the Bohr atomic model.

We have a single positively charged proton (black) forming the nucleus, and a single negatively charged electron (green) orbiting the nucleus.

The orbit distance n1 is defined by the electrostatic balance of the two opposing charges.

Andy Astbury,noise,light,Bohr atomic model

The Bohr Atomic Model

If we apply energy to the system then a ‘tipping point’ is reached and the electron is forced to move away from the nucleus – n2.

Apply even more energy and the system tips again and the electron is forced to move to an even higher energy level – n3.

Now here’s the fun bit – stop applying energy to the system.

As the system is no longer needing to cope with the excess energy it returns to its natural ‘ground’ state and the electron falls back to n1.

In the process the electron sheds the energy it has absorbed – the red squiggly bit – as a quantum, or packet, of electromagnetic energy.

This is basically how a flash gun works.

This ‘packet’ has a start and an end; the start happens as the electron begins its fall back to its ground state; and the end occurs once the electron arrives at n1 – therefore it can perhaps be tentatively thought of as being particulate in nature.

So now you know what Prof. Brian Cox knows – CERN here we come!

Right, so what’s this got to do with photography and camera sensor noise

Camera Sensor Noise

All camera sensors are effected by noise, and this noise comes in various guises:

Firstly, the ‘noise control’ sections of most processing software we use tend to break it down into two components; luminosity, or luminance noise; and colour noise.  Below is a rather crappy image that I’m using to illustrate what we might assume is the reality of noise:

Andy Astbury,noise

This shot shows both Colour & Luminance noise.
The insert shows the shot and the small white rectangle is the area we’re concentrating on.

Now let’s look at the two basic components: Firstly the LUMINANCE component

Andy Astbury,noise

Here we see the LUMINANCE noise component – colour & colour noise components have been removed for clarity.

Next, the COLOUR NOISE bit:

Andy Astbury,noise

The COLOUR NOISE component of the area we’re looking at. All luminance noise has been removed.

I must stress that the majority of colour noise you see in your files inside LR,ACR,CapOne,PS etc: is ‘demosaicing colour noise’, which occurs during the demosaic processes.

But the truth is, it’s not that simple.

Localised random colour errors are generated ‘on sensor’ due to the individual sensor characteristics as we’ll see in a moment, because noise, in truth, comes in various guises that collectively effect luminosity and colour:

Andy Astbury,noise

Shot Noise

This first type of noise is Shot Noise – called so because it’s basically an intrinsic part of the exposure, and is caused by photon flux in the light reflected by the subject/scene.

Remember – we see light in a different way to that of our camera. What we don’t notice is the fact that photon streams rise and fall in intensity – they ‘flux’ – these variations happen far too fast for our eyes to notice, but they do effect the sensor output.

On top of this ‘fluxing’ problem we have something more obvious to consider.

Lighter subjects reflect more light (more photons), darker subjects reflect less light (less photons).

Your exposure is always going to some sort of ‘average’, and so is only going to be ‘accurate’ for certain areas of the scene.

Lighter areas will be leaning towards over exposure; darker areas towards under exposure – your exposure can’t be perfect for all tones contained in the scene.

Tonal areas outside of the ‘average exposure perfection’ – especially the darker ones – may well contain more shot noise.

Shot noise is therefore quite regular in its distribution, but in certain areas it becomes irregular – so its often described as ‘pseudo random’ .

Andy Astbury,noise

Read Noise

Read Noise – now we come to a different category of noise completely.

The image is somewhat exaggerated so that you can see it, but basically this is a ‘zero light’ exposure; take a shot with the lens cap on and this is what happens!

What you can see here is the background sensor noise when you take any shot.

Certain photosites on the sensor are actually generating electrons even in the complete absence of light – seeing as they’re photo-voltaic they shouldn’t be doing this – but they do.

Added to this are AD Converter errors and general ‘system noise’ generated by the camera – so we can regard Read Noise as being like the background hiss, hum and rumble we can hear on a record deck when we turn the Dolby off.

Andy Astbury,noise

Thermal & Pattern Noise

In the same category as Read Noise are two other types of noise – thermal and pattern.

Both again have nothing to do with light falling on the sensor, as this too was shot under a duvet with the lens cap on – a 30 minute exposure at ISO 100 – not beyond stupid when you think of astro photography and star trail shots in particular.

You can see in the example that there are lighter and darker areas especially over towards the right side and top right corner – this is Thermal Noise.

During long exposures the sensor actually heats up, which in turn increases the response of photosites in those areas and causes them to release more electrons.

You can also see distinct vertical and some horizontal banding in the example image – this is pattern noise, yet another sensor noise signature.

Andy Astbury,noise

Under Exposure Noise – pretty much what most photographers think of when they hear the word “noise”.

Read Noise, Pattern Noise, Thermal Noise and to a degree Shot Noise all go together to form a ‘base line noise signature’ for your particular sensor, so when we put them all together and take a shot where we need to tweak the exposure in the shadow areas a little we get an overall Under Exposure Noise characteristic for our camera – which let’s not forget, contains other elements of  both luminance noise and colour noise components derived from the ISO settings we use.

All sensors have a base ISO – this can be thought of as the speed rating which yields the highest Dynamic Range (Dynamic Range falls with increasing ISO values, which is basically under exposure).

At this base ISO the levels of background noise generated by the sensor just being active (Pattern,Read & Thermal) will be at their lowest, and can be thought of as the ‘base noise’ of the sensor.

How visually apparent this base noise level is depends on what is called the Signal to Noise Ratio – the higher the S/N ratio the less you see the noise.

And what is it that gives us a high signal?

MORE Photons – that’s what..!

The more photons each photosite on the sensor can gather during the exposure then the more ‘masked’ will be any internal noise.

And how do we catch more photons?

By using a sensor with BIGGER photosites, a larger pixel pitch – that’s how.  And bigger photosites means LESS MEGAPIXELS – allow me to explain.

Buckets in the Rain A

Here we see a representation of various sized photosites from different sensors.

On the right is the photosite of a Nikon D3s – a massive ‘bucket’ for catching photons in – and 12Mp resolution.

Moving left we have another FX sensor photosite – the D3X at 24Mp, and then the crackpot D800 and it’s mental 36Mp tiny photosite  – can you tell I dislike the D800 yet? 

One the extreme left is the photosite from the 1.5x APS-C D7100 just for comparison.

Now cast your mind back to the start of this post where I said we could tentatively regard photons as particles – well, let’s imagine them as rain drops, and the photosites in the diagram above as different sized buckets.

Let’s put the buckets out in the back yard and let’s make the weather turn to rain:

Andy Astbury,Wildlife in Pixels,sensor resolution,megapixels,pixel pitch,base noise,signal to noise ratio

Various sizes of photosites catching photon rain.

Here it comes…

Andy Astbury,Wildlife in Pixels,sensor resolution,megapixels,pixel pitch,base noise,signal to noise ratio

It’s raining

OK – we’ve had 2 inches of rain in 10 seconds! Make it stop!

Andy Astbury,Wildlife in Pixels,sensor resolution,megapixels,pixel pitch,base noise,signal to noise ratio

All buckets have 2 inches of water in them, but which has caught the biggest volume of rain?

Thank God for that..

If we now get back to reality, we can liken the duration of the rain downpour as shutter speed, the rain drops themselves as photons falling on the sensor, and the consistency of water depth in each ‘bucket’ as a correct level of exposure.

Which bucket has the largest volume of water, or which photosite has captured the most photons – in other words which sensor has the highest S/N Ratio?   That’s right – the 12Mp D3s.

To put this into practical terms let’s consider the next diagram:

Andy Astbury,Wildlife in Pixels,sensor resolution,megapixels,pixel pitch,base noise,signal to noise ratio

Increased pixel pitch = Increased Signal to Noise Ratio

The importance of S/N ratio and its relevance to camera sensor noise can be seen clearly in the diagram above – but we are talking about base noise at native or base ISO.

If we now look at increasing the ISO speed we have a potential problem.

As I mentioned before, increasing ISO is basically UNDER EXPOSURE followed by in-camera “push processing” – now I’m showing my age..

Andy Astbury,noise,iso

The effect of increased ISO – in camera “push processing” automatically lift the exposure value to where the camera thinks it is supposed to be.

By under exposing the image we reduce the overall Signal to Noise Ratio, then the camera internals lift all the levels by a process of amplification – and this includes amplifying  the original level of base noise.

So now you know WHY and HOW your images look noisy at higher ISO’s – or so you’d think – again,  it’s not that simple; take the next two image crops for instance:

Andy Astbury, iso,noise,sensor noise

Kingfisher – ISO 3200 Nikon D4 – POOR LIGHT – Click for bigger view

Andy Astbury, iso,noise,sensor noise

Kingfisher – ISO 3200 Nikon D4 – GOOD LIGHT – CLICK for bigger view

If you click on the images (they’ll open up in new browser tabs) you’ll see that the noise from 3200 ISO on the D4 is a lot more apparent on the image taken in poor light than it is on the image taken in full sun.

You’ll also notice that in both cases the noise is less apparent in the high frequency detail (sharp high detail areas) and more apparent in areas of low frequency detail (blurred background).

So here’s “The Andy Approach” to noise and high ISO.

1. It’s not a good idea to use higher ISO settings just to combat poor light – in poor light everything looks like crap, and if it looks crap then the image will look even crappier.When I get in a poor light situation and I’m not faced with a “shot in a million” then I don’t take the shot.

2. There’s a big difference between poor light and low light that looks good – if that’s the case shoot as close to base ISO as you can get away with in terms of shutter speed.

3. I you shoot landscapes then shoot at base ISO at all times and use a tripod and remote release – make full use of your sensors dynamic range.

4. The Important One – don’t get hooked on megapixels and so-called sensor resolution – I’ve made thousands of landscape sales shot on a 12Mp D3 at 100 ISO. If you are compelled to have more megapixels buy a medium format camera which will generate a higher S/N Ratio because the photosites are larger.

5. If you shoot wildlife you’ll find that the necessity for full dynamic range decreases with angle of view/increasing focal length – using a 500mm lens you are looking at a very small section of what your eye can see, and tones contained within that small window will rarely occupy anywhere near the full camera dynamic range.

Under good light this will allow you to use a higher ISO in order to gain that crucial bit of extra shutter speed – remember, wildlife images tend to be at least 30 to 35% high frequency detail – noise will not be as apparent in these areas as it is in the background; hence to ubiquitous saying of  wildlife photographers “Watch your background at all times”.

Well, I think that’s enough to be going on with – but there’s oh so much more!

Become a patron from as little as $1 per month, and help me produce more free content.

Patrons gain access to a variety of FREE rewards, discounts and bonuses.