Noise is inherent to every sensor and at every ISO.
There are many different types of noise and many different reasons for noise. For example:
- Shot-noise or photon-noise. This type of anoise is a quantum effect (it's literally the nature of how the universe works and nothing you can do about it). It's the notion that light isn't a solid thing... it's the effect of individual photons randomly hitting locations. Even if you think you have perfectly even light, the truth is, randomly due to quantum effects, more photons will hit one sensor location than another sensor location. But the duration of most exposures is long enough that the statistical deviation from the mean is small. In other words this isn't the primary reason for the most of the noise, but it is a contributing factor.
- Dark-current / thermal signal. This is the notion that the warmer a sensor is, the more "noise" it will produce. In astrophotography where we routinely take exposures that are many minutes long (8 minutes... 15 minutes... 30 minutes... etc.) a lot of heat can build up in a sensor. So we have special cameras that chill the sensor to try to fight back this noise. For cameras that aren't hot and are taking very fast exposures, this noise is usually not a huge problem. But it is... yet another contributing factor.
- Bias noise. This type of noise is due to the nature of the sensitivity of the sensor. A sensor has to be powered up to work. If you kept the lens cap on the camera, powered up the sensor and took the shortest possible exposure (remember the lens cap is on ... so no light should be entering the camera), and then perform a read-out and analyze the pixels... you would notice the pixels (which should, in theory, all be black and thus have "0" values) in fact do NOT have zero values. This is the "bias" level. It's low... but... yet another contributing factor. (you can see how these "contributing factors" are all small... but starting to add up.)
- Hot & cold pixels... these are basically stuck pixels. It's yet another form of noise. Fortunately these can be "mapped out" (easily fixed automatically in post processing.)
- Read noise. This is the big ticket item and it is responsible for MOST of the noise you see in an image. This one is a bit more complicated to explain. Your sensor is NOT digital... it's analog (though there is a digital component). As photons hit the sensor, the "pixels" are building up a charge ... measured in electron volts (actually microvolts ... these are very low charges). This is like measuring the voltage of a car battery... it's an analog thing. But each pixel has some charge level to it. The trouble is... it's too weak to read. So this has to run through an analog amplifier to boost the voltage. That amplifier will NOT do a perfect job. Basically there's some wiggle room to the accuracy of the output from the amplifier. That signal is then passed to an analog to digital converter (ADC) and turn into a numeric (digital) value ... which is then stored on your memory card (to be interpreted into an image).
So about that noise...
Imagine, if you will, that you are going to photograph a piece of sky that is perfectly black... with the exception of one lonely star in the center.
Now imagine that you plot the brightness value of each pixel as a 3D graph.
You would expect the graph to be a perfectly flat surface (like water that is perfectly still) to represent all the pixels that have "0" value. And you would expect a few pixels at the center (where the star is located) to look like a large spike coming up from the surface.
But the reality is that you wont have all "0" pixels where "black" should be located. You'll have these very low values ... sort of like tiny waves on mostly flat, just not perfectly flat, water. And the nice big "spike" in the middle where the star is located.
If you don't boost ISO at all (base ISO or ISO 100 for your camera ... btw, ISO is really what we call "gain") you probably wont notice these tiny little "waves" on your surface (they are the "noise). Trouble happens when you boost ISO. This boost (really gain), amplifies the value of EVERYTHING in the shot. So the dim subject that you boosted by increasing ISO now no longer looks dim. But those really tiny waves that you could hardly notice... well they've turned into much bigger waves. They're now big enough to be noticed.
So what can YOU do about:
First, take a lesson here. ISO is _not_ part of the "exposure triangle". There are only two values in exposure... Aperture value, and Shutter value. That's it. I defend this because ISO is something that the camera applies ONLY AFTER the shutter has closed and the image is complete. It's a post-processing step to "amplify" what was captured at the end of the ACTUAL exposure.
The amount of noise you get is the amount of noise you get... ISO doesn't change the inherent noise. It changes the amount of amplification applied to EVERY pixel... causing the noise to suddenly be noticeable.
Anytime you are concerned about noise and have the freedom to choose a lower ISO, your best to pick the lowest workable ISO. Trouble is... in very low light, to get the version of the image where you don't notice the noise would require a very long exposure duration... and that would probably result in a blurry shot. So we boost the ISO because we can deal with noise in post processing more easily that we can deal with blur in post processing.
Onto the post processing...
You can use software to reduce noise. The software mostly works by calculating the average brightness (sampling clusters of pixels to determine how bright they are) and then knocking down the values of any pixels that stand out. But the problem is you will get strong (natural) deviation of pixel values in places that actually DO have loads of detail in the photo. And knocking down the differences means you're actually "softening" the shot... leaving you with something less-sharp than you want.
In some better de-noising algorithms, you can SELECTIVELY deal with de-noising vs sharpening (if you sharpen an image, you're asking software to accentuate the difference between pixels... making noise even MORE noticeable. If you de-noise an image, you're asking it to do the opposite... to minimize the difference between pixels... which does help hide the noise but also SOFTENS the image. Basically de-noising vs. sharpening are opposites of each other.
Also in better algorithms, you can ask the software to be more aggressive about noise in shadows than in bright areas. There are a couple of reasons why this is a good idea. One has to do with how noise works, the other has to do with how the human eye works (which we map to something we call "gamma").
Lightroom has some fairly decent algorithms and tools to deal with noise.
In Photoshop you can do some even more powerful things by learning to create an "edge" mask (the mask looks like someone converted the photo into an ink drawing... it identifies all the contrasty areas.) You basically duplicate the image, create the edge mask, and then de-noise the image with the mask. This causes all the NON-CONTRASTY areas to get de-noised, but leaves the edges alone -- remember that de-noising is the opposite of sharpening. So you don't want to de-noise edges of strong contrast or your image will look a bit soft.
In your image, you notice the noise in the violet areas on the walls. The mask would help you isolate the de-noising to just those areas.
Here's a pretty good video on how to create an edge mask... but this mask is used for sharpening and, of course, you want the opposite (de-noising). But you can invert the mask and use it to de-noise.
Anyway, there are more techniques (many more) on how to selectively deal with images to get rid of the noise only where it's needed and protect the details. This was just one example.
That's a lot to know about noise.
The key points are:
a) It's a reality of digital photography and ALL camers have noise and at ALL ISO levels (it's really just a matter of whether or not you can notice it).
b) There are tools and techniques that are pretty good at dealing with noise problems.
I often don't fuss over noisy images by creating edge masks in photoshop. Usually I start with the built-in tools in Lightroom. If that's not good enough, I use something like Imagenomic's "Noiseware Pro" or Nik "Dfine 2" (both of these are a bit smarter about dealing with noise) and call it a day.
But if I really wanted to fuss with noise & sharpeing for an image that was really important to me... then the technique of building an edge mask is a good one.
Tim Campbell
5D III, 5D IV, 60Da
