0
$\begingroup$

I want to remove the noise gaussian and poisson noise. But I'm unfamilier with noise remove methods but from my understanding it's similler to average filter which is median filter. But I want to know exactly how to remove those kind of noise.

My interest is in image. And I think that if I make an image with gassian noise +poisson noise then I have to use two remove stages. But I'm not sure whether can I use just one median filter to remove all noise or there are especially methds in there.

Improve this question
$\endgroup$
2
  • 1
    $\begingroup$ The answer will depend on what the noise is contaminating and what you want to do with the signal being contaminated. The methods applicable in digital communications (e.g. matched filtering) severely distort the signal and thus are not suitable for removing the noise from an audio signal that someone might want to listen to after the noise removal. -1 pending your edits that clarify what you really want to accomplish. $\endgroup$
    – Dilip Sarwate
    Jun 10 '15 at 16:45
  • $\begingroup$ @Dilip Sarwate My interest is in image. And I think that if I make an image with gassian noise +poisson noise then I have to use two remove stages. But I'm not sure whether can I use just one median filter to remove all noise or there are especially methds in there. $\endgroup$
    – gmotree
    Jun 10 '15 at 21:34
0
$\begingroup$

If you want to remove noise from an image corrupted by a mixture of Poisson-Gaussian noise (e.g. low photon counts lead to a Poisson noise component, and detector noise gives the Gaussian component) then there are couple of options:

  • Apply a variance-stabilizing transform such as the Anscombe Transform to essentially make the noise Gaussian, then use any technique for additive Gaussian noise removal. A good example is Optimal inversion of the Anscombe and Generalized Anscombe variance-stabilizing transformations, Mäkitalo & Foi (2014), paper and code

  • Work with the Poisson-Gaussian statistics directly. For example, Image Denoising in Mixed Poisson–Gaussian Noise, Luisier et al. (2011), paper and code are available.

Improve this answer
$\endgroup$
2
  • $\begingroup$ Should I have to treat it by separately? $\endgroup$
    – gmotree
    Jun 11 '15 at 13:01
  • $\begingroup$ @gmotree Both methods do a single treatment, but approach it in different ways. $\endgroup$
    – dr.blochwave
    Jun 11 '15 at 13:38

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.