Reverse image blur filter (not unsharp mask)

Question

I am looking for method to obtain reverse blur filter. I have original bitmap and I want to make something like Gaussian Blur Inverse.

Let us call it invB, and operation of Gaussian blur as function GBlur and original image O. Filter I would like to have gives invB as output, so that GBlur(invB) = O.

This operation has nothing to do with sharpening image (I assume it will disort image not make it sharper) and will never be used to restore original image from blurred one (it differs in assumptions and available data from that problem).

Making it short - Filter should make inverse of original image that will give original image after Gaussian blur. In InvB there are possibly more filters.

Thanks in advance.

#Edited.

Original:

enter image description here

After Sharpen and Gaussian Blur:

sharpenned and blurred rose

After Gaussian Blur then Sharpen:

blurred and sharpenned rose

So I have original image, I have to pass the image through Gaussian Blur ([[1,2,1],[2,4,2],[1,2,1]], / 16) and prepare original image before blurring, so that it will be equal to original image (whenever possible, and closest to the original image when not possible).

Maybe because of the problem nature I cannot find anything via Google. Please, do not tackle this problem by deblurring - it simply does not help.

can you describe more about your problem? What's your dataset? There are many deblurring algorithms out there — azer89
Have you Googled? There are plenty of hits - for example this one and this one and this one. Does your problem differ from these? — Roger Rowland
"This operation has nothing to do with sharpening image": whether you like it or not, deblurring is sharpening and nothing else. — Yves Daoust

Yves Daoust Yves Daoust · Accepted Answer · 2014-02-15T16:54:36

What you describe is indeed a deblurring filter, whether you apply it to a blurred image or not. It can be implemented by inverse convolution, also called deconvolution. In the Fourier domain, it amounts to dividing by the (Gaussian) filter response instead of multiplying.

Have a look at the Richardson-Lucy method http://en.wikipedia.org/wiki/Richardson%E2%80%93Lucy_deconvolution.

In practice the results are often quite disappointing because of strong noise enhancement and ringing. To reduce these effects, the Wiener filter is preferred http://en.wikipedia.org/wiki/Wiener_deconvolution.

Reverse image blur filter (not unsharp mask)

2 Answers