5
$\begingroup$

If we convolve an image with a point spread function and from the resulting image to find the input image, can we use any stochastic approaches? I feel like we will not be able to. A single image seems to me a deterministic quantity and I cannot think of any way to approach this deconvolution problem in a stochastic way. However, I am not sure and I want to know if there is a way. Any help is appreciated

$\endgroup$
5
  • $\begingroup$ Any approach which takes into account the noise in the image is stochastic. $\endgroup$
    – Royi
    Jan 17 '21 at 19:38
  • $\begingroup$ I see, thank you so much @Royi ! $\endgroup$
    – xhensa
    Jan 18 '21 at 9:13
  • $\begingroup$ Would you accept this as an answer? $\endgroup$
    – Royi
    Jan 18 '21 at 9:22
  • $\begingroup$ Though I could not figure out how it would work in the real case, yes, I would accept it as an answer. But if you are asking for me to accept your comment as the approved answer, I would like to say that you have written it as a comment not an answer, hence I am not able to accept it as the answer $\endgroup$
    – xhensa
    Jan 19 '21 at 12:59
  • $\begingroup$ Hi, I added it as an answer. $\endgroup$
    – Royi
    Jan 19 '21 at 15:48
3
$\begingroup$

Any Deconvolution method which takes into account the noise in the image is basically a stochastic approach.

Usually, the model for Deconvolution is:

enter image description here

So having the noise in there makes it a problem with stochastic properties.

Remark
If by stochastic you meant sampling from the Posterior Distribution then you may have a look at Stochastic Image Denoising by Sampling from the Posterior Distribution (Though it is not about Deconvolution but on Denoising).

$\endgroup$
1
  • 1
    $\begingroup$ Thank you for the nice explanation! $\endgroup$
    – xhensa
    Jan 19 '21 at 15:53

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.