1
$\begingroup$

I am dealing with an image restoration problem with noisy measurement. I use the classical total variation norm to remove the noise, specifically, the TV-$L_2$ problem. I used the famous Lena image for testing (size $256\times 256)$.

enter image description here

The result is attached below:

enter image description here

As can be seen, there are some dot-patterns like noise still existing in the denoise image. When I increase the coefficient for TV norm, the whole image becomes blurred and details are lost.

Could anyone tell me how to deal with these local non-uniform noise patterns? It seems that they are bandpass patterns.


P.S. I am dealing with an image reconstruction problem named phase retrieval with noisy measurements. The measurements are collected via practical devices. I use the TV norm as the regularization term to improve the reconstruction performance. However, the non-uniform artifacts appear in the reconstructed image (see above). So are there some denoising algorithms that could deal with this situation? I adopt an optimization framework that can treat the proximal operator as a denoising problem. So I just focus on the denoising problem.

$\endgroup$
13
  • $\begingroup$ Are you adding noise to the image or trying to denoise it as is? $\endgroup$
    – Royi
    Dec 1, 2021 at 12:47
  • $\begingroup$ @Royi The measurement is not from the simulation but from the practical device. I assume that the Poisson noise, Gaussian noise, and saturation situation exist. $\endgroup$ Dec 2, 2021 at 1:59
  • $\begingroup$ @Royi Specifically, I deal with the problem named phase retrieval. I collected the Fourier intensity measurements and reconstruct the original image. The measurement is acquired by the CMOS camera. $\endgroup$ Dec 2, 2021 at 2:00
  • $\begingroup$ Are you after a TV Denoising with less artifacts? $\endgroup$
    – Royi
    Dec 2, 2021 at 4:37
  • $\begingroup$ @Royi Hi, I use TV denoising, the whole image becomes smoother (sharp details lost) although with less dot-like patterns. $\endgroup$ Dec 2, 2021 at 9:33

1 Answer 1

0
$\begingroup$

Your noise model is similar to the noise model of SAR which is basically a phase reconstruction. You may look at this survey: Image denoising and despeckling methods for SAR images to improve image enhancement performance: a survey.

$\endgroup$
1
  • $\begingroup$ A $40 article? It better be good! $\endgroup$ Aug 14 at 19:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

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