Deconvreg 2D deconvolution gives too large result

Hi everyone,

I'm trying to deconvolute a 2D image recorded with scanning microscopy with an estimated PSF. Since both the PSF and the final image are 2D I'm doing a line by line deconvolution, basically scanning the PSF through the image via for loops in order to receive a 2D deconvoluted image.

The resulting image is pretty much what I'd expect however the values are too big by a factor of about 1000. My PSF, while only an estimate, is unlikely to be off by that much and playing with its shape and amplitude has only resulted in a factor 10 change.

Currently my suspicion is that it is because the two images I'm trying to deconvolute are physical quantities, that is they are fields sampled in space. I reckon I probably have to take into account the units (square meters) which I believe deconvreg does not do. Do you think that assumption is valid and if so do you know how I can normalize deconvreg to achieve a better result?

Thanks for the help!

Answer 1

[Matt J]

Try multiplying by the pixel area.

[Achim H]

Hi Matt,

thanks for your answer and sorry for not replying earlier, I thought I would get an email notification. How would multiplying by the pixel area help? If I understand it correctly, this would in my case actually increase the factor I am off by. Since the number of pixels defines the resolution of my scan wouldn't this also be somewhat arbitrary?

[Matt J]

As you say, the deconvolution has no knowledge of your spatial dimensions, so integral transforms like convolutions/deconvolutions will be approximated by discrete sums. You need to multiply the sums by the physical area of a single pixel (in square millimeters or whatever the appropriate units are) in order to give them the same quantitative scale as integrals.

[Achim H]

Awesome thanks, that makes sense, I figured it was something related to the units but did not quite know how to account for that. I also found this explanation on the Gwyddion help page that basically states what you just said.

Thanks!