|
Common Image Problems: Noise
Noise is one of the most hotly discussed subjects on photography forums. Which sensors create more noise? How much noise reduction should I use? Which ISO settings should I use? Is my 12MP sensor more noisy that your 10MP sensor? etc.
In this section we will look at what causes noise, how you can minimise it, and what you can do to reduce the effects afterwards.
What do we mean by noise?
I think a series of pictures will illustrate this better than anything (this image is a crop from a much larger image and shown at 1:1 zoom) ...
(images below used courtesy of www.imaging-resource.com - an excellent website for detailed comparisons of a huge range of cameras).
 |
A high end DSLR camera - image taken at a low ISO setting to minimise the amount of noise. |
 |
This is taken with an entry-level DSLR at an ISO setting of 1600. Still remarkably good (remember we are looking here at individual pixels and a 7x5 print made from this image would show no noise).
Noise is starting to show up in darker (shadow) regions of the image - particularly in areas of constant colour such as the colour card. |
 |
This is also a 1600 ISO image taken with a leading compact camera. The noise is seen everywhere and this would clearly show up in a print.
The manufacturer of this camera has not applied as much noise reduction as most do so although the image is noisy, there is still a reasonable amount of detail. |
 |
I kid you not ... this is an image from a brand new fairly high end compact camera.
In the ridiculous race of numbers going on between the leading manufacturers so they can quote the highest megapixels and highest ISO settings, this picture was taken with a Fuji F100fd - a 12MP camera - with the ISO set to 12800 !!!
There has been so much noise reduction applied to this image that even the coloured square are no longer square.
However without the noise reduction - you would have probably not even be able to recognise the image with random noise making up most of the captured detail. |
Pixel Peeping
This is a term used to describe people that look at images at 1:1 zoom (actual size) so that they can see the individual pixels and therefore talk/debate with anyone who will listen that their camera has less loise than someone elses.
Pixel peeping is both a good and a bad trait. It is the only way to truely see the noise levels in your image so if you are looking at applying any sort of noise reduction or filtering, you do need to view the image at 1:1 size.
However doing this can give you a false impression of how noisy the image is. If you want to make a 6x4 print from a noisy image (maybe one taken at 1600 ISO which looks covered in noise when viewed 1:1), you will probably be very surprised at how little noise shows up on paper. Try it and see for yourself.
What causes noise?
The primary source of noise is the image sensor. Every electronic device will produce noise (unless you can get it to operate at a temperature of absolute zero) - it is the equivalent of background hiss on hi-fi equipment. When the amount of light falling on a pixel is read off, the actual value will have some random variation - this is called noise.
The amount of noise generated by the sensor is called the 'nosie floor' - and to get a decent quality image, you need the actual signal from the pixels to be significantly greater than this. This difference is called the signal-to-noise ratio.
In bright light conditions, the signal from the sensor is well above the noise floor so you have a good signal to noise ratio. You therefore don't need to set the ISO too high to amplify the signal so the noise remains very low in the final image.
In low light however, the strength of the signal is much lower and therefore closer to the noise floor (a low signal-to-noise ratio). To get a bright image, you need to amplify the signal by increasing the ISO setting. However amplifying a signal also amplifies the noise so you are not actually improving the signal-to-noise ratio, just brightning the image and the noise together. This obviously causes it to show up more in your image.
The key to low noise images is to get the greatest signal-to-noise ratio - and the only way to do this for a given camera is to have sufficient light to put the signal sufficiently above the noise level (and therefore allowing you to use a low ISO).
Larger image sensors generally have much larger individual pixels, and therefore the ability to capture more light. The noise level is independent of the pixel size, so therefore a bigger sensor leads to a better signal-to-noise ratio and lower noise images. This is why a DSLR camera will always produce much lower noise images than a compact camera.
Compact cameras often have a range of ISOs from 100-1600 (although some manufacturers are putting ridiculously high settings in the hope to win customers - see images above). Once you get much past about 400 however, the image quality can deteriorate very rapidly. This happens in one of two ways depending on the camera's built in software. Either the image can become very noisy or the image processor can apply ever more severe forms of noise reduction algorithms which will remove the noise, but also most of the finer detail in the image.
An extra note about long exposures - noise increases with temperature, and when you make a long exposure, the temperature of the sensor increases. Therefore the following two exposure should give an equivalent image brightness, but the second one would probably show more noise.
- ISO 100, f/4, 1 second
- ISO 100, f/22, 30 seconds
In addition to the general increase in noise caused by temperature, long exposures can also reveal that certain pixels may stand out against the surrounding ones and output a much greater noise signal.
Some DSLR cameras have special noise reduction algorithms which kick-in only during long exposures. These often work by taking a second exposure of the same duration as the main exposure, but not actually opening the shutter - therefore the sensor only records the sensor noise. It can then subtract this from the original image.
So what can be done about noise?
There's one simple answer to this - keep your ISO setting as low as possible. Of course things are never that simple. If the light levels are low, leaving your camera on ISO 100 may lead to a shaky shot caused by a slow shutter speed - so the sensible thing to do is to sacrifice a little noise to get a steady image.
It pays to know how well your camera performs at the different ISO settings so you can have in your mind how high you are willing to go. I used to use a Fuji S9600 which was a great camera provided it was kept at ISO 400 or below. ISO 800 was just about usable, and ISO 1600 looked like a water colour painting due to severe noise reduction.
If you have a DSLR, you can usually go up to the maximum of 1600 (or 3200 on some) without the image quality being too badly effected by noise, but with compact cameras, noise starts to show up much sooner. So know how your camera performs at 400, 800 and 1600 ISO, and anything beyond that, forget it.
Noise Reduction
Noise Reduction refers to mathematical algorithms that attempt to reduce / remove noise from an image without actually removing any of the original image detail. It achieves this by knowing certain characteristics of noise that might appear, looking for these and removing them. The hope of course is that there is nothing in the original image that has any of the characteristics of noise otherwise these will be removed too.
The reality is that noise is random and can easily be confused with fine detail in an image. Colour noise is random specks of colour (very noticable in shadow areas of an image) and to reduce this means that the overall colour saturation is reduced. Look how dull the colours are in the fourth image at the top of this page.
If you shoot in JPEG format, then all the noise reduction is done by the image processor on the camera before the file is created (although you may have some control through the cameras menu system as to how much noise reduction is applied).
If you shoot in RAW however, no noise reduction is applied and you can control this in post processing on your computer. You can then decide how much noise reduction to apply - reducing more of the noise but also losing detail in the image. All RAW image processing software will have some noise reduction capability, although the quality of this varies greatly. Adobe Camera Raw seems to do a reasonable job, however there are dedicated packages out there that do a much better job. One of the more widely respected ones is called Noise Ninja (http://www.picturecode.com/) - and this might be worth looking into if you shoot a lot of high ISO photography. Other's available include Neat Image and Dfine.
|
