I want to implement Wiener filter. I have original image and image blurred with linear motion. The one thing I am lack of(to implement it) is Noise power spectrum.

I know that it is hard task. I've took a look on different papers:

I don't understand why it is that complicated to just find noise spectrum if I have original signal(image)?

If I take noise image minus original image, it will be my error image. And if I estimate variance of it, that will be my Noise variance(Noise Power)?

I totally confused and here is my question: " How to estimate Power spectrum of the noise of the blurred image if I have original image?"

Btw, I don't want get some matlab function which will do all of it, but rather algorithm or equation that I can implement.

  • $\begingroup$ Oh, really! I'm sorry, I didn't know that you have this! Thank you! $\endgroup$
    – Mykola Servetnyk
    Commented Jan 7, 2015 at 1:15
  • $\begingroup$ No worries! :-) $\endgroup$
    – A. Donda
    Commented Jan 7, 2015 at 12:22
  • $\begingroup$ Could you review my answer? $\endgroup$
    – Royi
    Commented Jul 30, 2023 at 6:27

1 Answer 1


Usually the model is something like:

$$ \boldsymbol{y} = \boldsymbol{H} \boldsymbol{x} + \boldsymbol{n} $$

Where $ \boldsymbol{y} $ is the output image, $ \boldsymbol{x} $ is the input image, $ \boldsymbol{H} $ is the operator (Blurring, Decimation, etc...) model and $ \boldsymbol{n} $ is the added noise.

Indeed it is easy to know the spectrum of $ \boldsymbol{n} $ if you have both $ \boldsymbol{y} $ and $ \boldsymbol{x} $ (And $ \boldsymbol{H} $).
Yet in practice, the model means we only have $ \boldsymbol{y} $ and some prior on $ \boldsymbol{H} $ and $ \boldsymbol{n} $.

Then the problem is not trivial at all.
It is called Inverse Problem.


Your Answer

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

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