All About Noise

Want noise?  Shot under studio lighting with Nikon's D5.  It's a bit noisy, but that is because of the amount of voltage driven to the sensor to work at ISO 409,600   Given that incredible ISO setting, I would say that the image is quite awesome.

Want noise?  Shot under studio lighting with Nikon's D5.  It's a bit noisy, but that is because of the amount of voltage driven to the sensor to work at ISO 409,600   Given that incredible ISO setting, I would say that the image is quite awesome.

As you may know, I provide support and moderation on some different photography forums, and as often comes around, the subject of grain and noise has appeared again, and as is common, the subject is accompanied by an ocean going freighter filled with bullcrap, pseudo science and completely wrong so-called "facts".

Before we go too far and someone who doesn't read the entire essay presumes that the Nikon D5 is noisy, here's the same scenario at ISO 100

Same shot but taken at ISO 100.  The amount of voltage to the sensor has huge impact on the amount of digital noise

Same shot but taken at ISO 100.  The amount of voltage to the sensor has huge impact on the amount of digital noise

First let's get our definitions in place.  Digital files have no grain.  Grain is an outcome of the chemical process that creates photographic emulsions laid on safety film.  There is no chemical silver halide crystal based emulsion in digital so there can be no grain.  You can apply a digital effect called grain that emulates the appearance of grain, but natively there is no such thing as grain.  It's digital noise.  You can call it grain, call it ferdibots for all I care, but in the spirit of being accurate, let's agree that it is digital noise.

What Causes Digital Noise

Digital noise is the result of the application of voltage to a photosite.  Like analog noise in the old world of stereo, it is a by product of inefficiencies in the electronics.  Our digital sensors capture photons in what is commonly called a photosite, and sometimes referred to as a pixel.  To increase the ability of a photosite to collect photons, we increase the voltage to the photosite.  When we do this, we decrease the ratio between good signal and bad noise.  This is called the signal to noise ratio.  In a perfectly efficient world, any applied voltage would result in 100% signal and 0% noise at any applied voltage.  That's not our world so there is always noise, but as voltage increases, so does noise.


A cusp is a transition zone, or point of inflection.  Signal to noise rations do not change linearly, they act on a curve.  I will save you the engineering explanation, partly because it is long and a bit tedious and partly because I do not trust my ability to make it readily understandable.  This is why on our digital cameras, we will make decisions that say, "the camera is great to ISO 3200, but anything higher is really noisy".  Every time we jump an EV, we are doubling the sensitivity of the sensor, but this does not always equate to a doubling of the noise.  Sensor makers work hard to move the inflection points higher and higher up the curve, so an older sensor will be noisier at a given ISO than a new sensor.  Probably, but not always.

The Megapixel Lie

We know that the uninformed chase megapixel count over all else.  More megapixels should mean higher resolution and a larger native print size, but that also means that the physical area of each photosite is much smaller, so to gather the same amount of light in a smaller area, we have to push more voltage to it and that means that we get more noise.   There is a very good reason that Sony made to a7 Mark II specialty bodies.  The a7S Mark II had only 12 megapixels, but because the photosites were so large, they could be super efficient and are nearly noise free at higher ISOs.  The a7R Mark II had over 42 megapixels, which made the resolution of the sensor wonderful, but at the price of much higher levels of noise at every ISO because more voltage had to be put to each photosite to gather light.  As photographers, one of the many balancing acts we perform is choosing resolution over noise, or the other way around.

Sensor Size Matters

A smaller sensor has smaller photosites given a near equivalence in megapixels.  In the example that will follow, I compare a Canon 5D Mark III and a Canon 7D Mark II sensor.  The 5DIII is a full frame sensor, and the 7DII is a crop sensor.  Yet they have very comparable megapixel counts, thus the 7D II has a much smaller area per photosite, so to gather light at the same ISO we need to push more voltage to each photosite, and this will make the smaller sensor noisier.  And it is.  Whether I think that the 7D II sensor is good, is immaterial, what matters is understanding it's strengths and limitations, and one limitation is that it is noisier than a full frame sensor with similar photosite count.

RAW Rules

JPEGs as we know, are processed, and part of the processing that happens in camera is some form of noise reduction as well as sharpening.  We see JPEGs on the rear LCD and sometimes wonder why the RAW looks less sharp and more noisy.  It is simply because it is.  However because even JPEG Fine throws away over 70% of the total capture data, a JPEG has a lot less post processing latitude, so for the best in noise management, we should be shooting in RAW.  This is not an opinion, it's a mathematical fact. 

I am developing another essay on the nature of RAW files from one vendor to another.  While a RAW is ostensibly the uncooked data, there is processing that goes on before the RAW file gets created and some manufacturers do a lot more to the RAWs than others.  It's kind of like eggs.  The ones that we buy at the supermarket probably all come from chickens, but some chickens are treated differently than other chickens which ostensibly makes some eggs better than others, or at the very least makes them different.

You should still shoot in RAW, just note that your camera's RAW may not be as RAW as they might have you believe.

A Canon Example

For the purposes of conversation, let's look at two sensors from Canon.  The first will be from the 5D Mark III and the second from the 7D Mark II.  I chose these, because many photographers own them as a shooting pair, because they are equivalent pretty much in resolution, and they have two different sensor sizes.

Camera Model Sensor Size (mm) Megapixels Resolution Pixel Pitch (microns) Pixel Area (microns squared)
Canon 5D Mark III 36x24 22.3 5784x3846 6.22 38.69
Canon 7D Mark II 22.4x15 20.2 5486x3482 4.08 16.65

What we see here is that while the resolution is pretty close, the distance between photosites (pitch) is much less and the total area of the photosite is less than half on the 7D Mark II compared to the 5D Mark II.  Thus we have mathematical proof of much smaller photosites.

This tells us, that on the crop sensor 7D Mark II we are going to be using more voltage at each photosite to gather the same amount of light as on the 5D Mark III because the photosites are smaller and so need more power to be equivalently sensitive.  From this we can conclude correctly that the APS-C sensor will be noisier at any given ISO, and we may also determine by testing that the inflection point will be a lower ISO than on the 5D  Mark III.  And it is.

Sensor Tonal Range

Number of Available Tones per bit range (14 Bit RAW) Grey Scale Maximum RGB Value at the bit value (approx)
1 18
2 36
4 55
8 73
16 91
32 109
64 128
128 146
256 164
512 182
1024 200
2048 219
4096 237
8192 255

As I have explained in my essays on how sensors capture data and why Expose to the Right makes so much sense for photographer's, a camera sensor does not record the same range of tones at each level of bit depth.  For example, a sensor with a bit depth of 14, also designated as capable of producing a 14 bit RAW file, has far fewer tonal options in the darker bit areas than in the light.  The number of tonal variants pretty much doubles with each tone area increase.  This is sometimes confusing to folks but starts to make sense when you think about how 2 to the power of 14 is done.  You start with 2 and for each iteration, you multiply the previous product by two again, 14 times.  Each value you get in the process is a decent indicator of the number of tonal variances in each block of the multiplication process.

To help make this make a bit more sense, I've added the very approximate RGB grey scale value for each factor.  Since 2^14 is 16384, we know that there are a total of 16384 potential tonal values in the file, but as we see from the sensor math, more tonal values are available the closer that we get to the RGB group that tops out at white.  Each approximate RGB value starts where the last one stops and goes to the approximate number shown.  For example the range at 16 goes from RGB 73 to RGB 91 in 16 steps.  The math may look a bit wonky, but it is a standard exponential expansion.

What this means for most of us, is that we will be capturing a wider range of tones and information the more exposure we capture, up to the point where clipping happens.  This is why expose to the right works.  Yes the image out of camera will be overly bright, but when you adjust the luminance in post down, you are actually moving more tonal variances into an area that could never naturally have that many.  

Back in the days of film, we used to say to underexpose slightly to increase the colour saturation and pop.  I still hear people say that about digital.  It may look good on the small LCD, but it fails completely mathematically because it is counter to how sensors actually work.

Noise and Long Exposure

The longer the shutter is held open, the more voltage is being delivered to the sensor, so the noisier the image gets.  This is one reason why long exposures should be done at lower ISOs to avoid compounding the problem.  There is a function in most cameras called long exposure noise reduction to help reduce noise in long exposures. 

What happens is that after the primary exposure is made, the camera makes another one of the same duration that is all black and then combines the two together to drop out noise.  This happens in RAW and in JPEG so this is one example where RAW is not as RAW as we might think.

Noise Myths

I have heard and read people say that different lenses create more noise.  I have heard people say that neutral density filters create noise.  I have read it that polarizers create noise.  This is all BS.  Noise is created by voltage not glass, anymore than grain had anything to do with glass.

If the exposure is underexposed in an area of importance and the underexposed area is brightened, that will create the appearance of noise, not because of more noise, but because of less data in darker areas.  Read my article on Exposing to the Right to see charts of how much less information is recorded in the darks compared to the lights in the same exposure and this will all make sense.

Thus with darkening filters and with slow lenses, we must be diligent in getting the exposure correct, and it is important to understand the limits o the built in light meter when the light it measures is very low, such as one may find at night or with a high density Neutral Density filter.  Particularly for high value Neutral Density filters, I recommend using a smartphone app such as NDTimer to determine the correct exposure change for a given ND density.

Viewing Distance

Do also consider where the image will be shown and at what viewing distance.  Many of us have larger displays, such as the popular 24" or 27".  Proper viewing distance from these displays is more than four feet for image viewing, but we all tend to sit about a foot away, so everything is more evident.  Then we edit at 100% and everything looks horrible.  Take a step back before freaking out about the noise and then overprocessing and ruining the image.


Digital noise is a natural outcome of digital photography.  All images will have some noise or other.  If it really bugs you, consider the use of a post processing plugin to clean it up after the fact.  There are many, including those built into Lightroom.  My personal favourite is Dfine found in the Nik Collection, but Topaz DeNoise does a great job as do others.

Lastly, please remember these words from instructor and mentor Rick Sammon.  "When you look at an image and all you see is noise, it's a bad photo."  When folks get caught up on noise, I suggest that they look at Migrant Mother by Dorothea Lange and tell me what they think.  19 times out of 20 no one mentions that this photograph is extremely grainy.  If the photo is great, the noise does not matter.

Have you heard other folk tales about noise?  Share them with us by leaving us a comment.  I will be sure to provide the straight goods on any of them if you ask.

If you shop with B&H Photo Video, please consider doing so through the link on as this helps support my efforts and has no negative impact whatsoever on your shopping experience.  If you find the podcast or articles of value, consider clicking the Donation tab in the sidebar of the website and buy me a coffee.  Your donation goes to help me keep things going.  Email your questions on any photo or video topic and I will try to respond within a day.

I'm Ross Chevalier, thanks for reading, and until next time, peace.