Introduction to digital noise

Back in the film days, high ISO films were notorious for creating a very grainy image. It was the trade off for having enough sensitivity to light to be able to capture low light situations. In today's digital age we have a similar problem called digital noise or image noise.

Image noise is something most serious photographers have to take into account when creating a photo. Specially if a paying client is going to be on the receiving end. No one wants a picture that looks grainy with a lot of imperfections. Casual shooters could probably care less though I find many of them worry about it simply because it's mentioned so often everywhere you read.

Image noise is the uneven distribution of pixel tones when compared to adjacent pixels in a given field of similar tones. Meaning, if you take a photograph of a smooth colored surface, that surface tone should transition across with very little fluctuation of extreme tonal changes from pixel to pixel. I probably lost you there.

If you look at the first sample on the left you will see a smooth gray blend with a slight tonal shift from dark to light as you look from left to right. The close up section shows a nice even distribution of tones from pixel to pixel, creating a smooth transition.

The second sample has that same transition but with a heavy noise ratio (exaggerated for this example). You can clearly see the extreme fluctuation of pixel tones where adjacent pixels do not transition smoothly from one to the next. There are darker pixels next to lighter pixels with no order. This is digital noise or image noise.

Image noise can originate in two areas; in the camera and in the editing.

In the camera: As we already established, higher ISO ranges are susceptible to higher noise ratios. This is caused by uneven signal amplitude on individual receptor sites on the digital image sensor. Since we are talking about a sensor with several thousand receptor sites all working independently of each other, the probability of each site sampling light at the same rate is almost impossible to control. Think of it like having a thousand adults with latex balloons and each person has three seconds to inflate their balloon. Some will have slightly more or less air than their neighbor. A silly example but one that illustrates what is happening on a digital sensor.

High ISO isn't the only culprit. You can get increased digital noise during long exposure times with a low ISO. Using our example let's tell those thousand adults they have five minutes to blow up their balloon and ask them to take their time. Some people will misjudge and blow too fast while others not fast enough. That translates to pixels that are too bright and too dark respectively.

A final issue that can affect image sensor noise is the physical temperature of the sensor. Sensors are delicate pieces of electronics susceptible to temperature fluctuations. Increased internal temperatures in warm weather will definitely affect the sensor.

In editing: Not wanting to get too technical, luminance range capture on a digital sensor is not linear. There is a sloping, or tapering, on the deep shadow and bright highlight sides of the range. It's not as sensitive as film, but it's there and can make for some interesting reading for you technophiles. In layman's terms it means that data on the dark shadow end and the light tones end have a short working range before they clip to black or white. Take that data and stretch it out (extend the range) by using curves, shadow or highlight recovery and you'll begin to notice noise. More so on the shadow end.

Pushing luminance ranges up past acceptable levels exaggerates the differences in tones between neighboring pixels. When they are in their native state that difference is almost imperceptible. Shifting them or pushing them in an editor artificially amplifies the signal and makes the imperfections more obvious. That's not an easy example so let's revisit our group with their balloons.

I give my thousand people three seconds to fill up their balloons. They will represent the shadow end of the dynamic range. In those three seconds most people get one breath in while some get two breaths in. We also have a few experts who got three breaths in. Looking at these thousand balloons with one, two and three breaths there's not too much difference. I then take my giant breath maximizer machine and it mechanically quadruples the number of breaths in the balloons. Suddenly I have balloons with four, eight and twelve breaths of air in them. Those with twelve breaths look way bigger and stand out more among the crowd than their four breath neighbors.

While digital noise is not something everyone has to worry about it is still wise to understand what causes it. By understanding this you can avoid or limit certain practices that will help you minimize digital noise. I know these are hockey examples but hopefully they help illustrate, in simple terms, how noise works.