1
$\begingroup$

Recently I read about image noise. I read about noise characteristics, but I am unable to understand the following two points:

The sources of noise in digital images arise during image acquisition (digitization) and transmission.

  • Imaging sensors can be affected by ambient conditions

  • Interference can be added to an image during transmission

$\endgroup$
2
$\begingroup$

If you assume the "ideal" image is what you see with your eyes (okay, this is a bit of a stretch, none of our eyes are perfect), then you can classify the noise based on where or how the differences were introduced.

Noise caused by sensors (some examples of this could be:)

  • if you aim your camera at the sun, you'll just have a very bright spot and "flash" on your image (which you do not see with your eyes)
  • you've seen this when trying to take an photo of a starry sky: your eyes can see the stars perfectly, but most cameras will just produce dark/black images with no starts
  • some modern sensors have an in-build brightness normalization (e.g. my laptop camera): if taking a picture/video with a bright object in the background (e.g. sitting in front of a light or a window when video-chatting) can sometimes make the foreground object (e.g. the person) overly dark
  • the camera lens has it's intrinsic parameters which introduce distortions in comparison to what we would expect
  • anything related to the environment and camera characteristics (more good examples on wiki)

Noise introduced "digitally" (I would say that transmission is just one example):

  • if an image is being transmitted between two computers on the network, losses/errors can occur. Especially when dealing with transmission with no error detection/correction, some bits can change. If you consider the image that was sent as the ground truth, the changed bits will be noise in the received image.
  • another example could be lossy image compression. A simple example would be if you decided to represent your pixels by 4-bits instead of 8-bits of grey levels. And suddenly you only have 16 gray levels available instead of 255, and noise gets introduced.
  • any noise (digitally)introduced to the image that makes it different from what was initially acquired by the sensor would go to this category.
$\endgroup$
4
$\begingroup$

Imaging sensors are just electronic devices that receive photons (light) and turn them in to electrons (electric signal). As such, they are affected by ambient conditions such as temperature. For example, you will have some noise coming from the electronic agitation (false currents due to freely moving electrons, without incoming light) that grows with the sensor temperature.

You do also have non-linear responses in the sensor that cause other noise patterns. A pixel needs to receive enough light before providing a reliable measurement. Before that threshold, it will be affected by Poisson noise (a noise whose distribution depends on the value of the pixel). On the other side of the per-pixel energy, it can get saturated (too much light), and its energy will "spread" onto neighboring pixels, thus modifying their value.

Finally, since you have a digital device, then you have an analog-to-digital conversion step that is likely to introduce further quantization noise, i.e., the recorded values will not be the "real" ones but something close up to the measurement precision.

Outside your imaging device, there is light transmission in the atmosphere (turbulence is very likely to be a disturbance cause here, but it's usually not considered as noise), and image transmission over communication channels where packet losses (and thus pixel loss!) can happen.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.