Deconvolution

pwnell · Post by **pwnell** » Thu Aug 15, 2013 5:13 pm

My first attempt at 3D deconvolution.

Original stack of 28 images using HF and Photoshop:

Deconvolution using cellSens Constrained Iterative:

René · Post by **René** » Fri Aug 16, 2013 2:16 am

Wow! Great result. Can you show us a single image of your stack, for a comparison??

CellSens, is that from Olympus? Is there freeware that you know of?

Best wishes, René

RogelioMoreno · Post by **RogelioMoreno** » Fri Aug 16, 2013 4:28 am

Waldo,

Amazing results!

Do you have any idea the price of cellSens Dimension?

Rogelio

Cyclops · Post by **Cyclops** » Fri Aug 16, 2013 8:43 am

What on earth is deconvolution?

pwnell · Post by **pwnell** » Fri Aug 16, 2013 9:34 am

René wrote:Wow! Great result. Can you show us a single image of your stack, for a comparison??

Here:

René wrote:CellSens, is that from Olympus? Is there freeware that you know of?

Yep, from Olympus. No, it is quite expensive. I have their 30-day trial and am trying to assess whether it is worth while.

pwnell · Post by **pwnell** » Fri Aug 16, 2013 9:35 am

RogelioMoreno wrote:Amazing results!

Do you have any idea the price of cellSens Dimension?

Thanks. Not exactly sure, but I do know it is a couple of thousand dollars.

pwnell · Post by **pwnell** » Fri Aug 16, 2013 9:38 am

Cyclops wrote:What on earth is deconvolution?

It is a process for reversing optical distortion. In theory, if you have all the parameters of a perfect lens and the other variables you can reverse optical distortions by calculating a function (PSF) that explains what happens to a point source of light as it travels from subject to the CCD.

Cyclops · Post by **Cyclops** » Fri Aug 16, 2013 11:29 am

pwnell wrote:
Cyclops wrote:What on earth is deconvolution?
It is a process for reversing optical distortion. In theory, if you have all the parameters of a perfect lens and the other variables you can reverse optical distortions by calculating a function (PSF) that explains what happens to a point source of light as it travels from subject to the CCD.

I see, I think. Do you have an example in "normal" light?

pwnell · Post by **pwnell** » Fri Aug 16, 2013 11:35 am

Well the first image is in fluorescence before deconvolution. So that would be "normal" light I guess? I have posted a reflected brightfield image in another thread on the fern stomata.

discomorphella · Post by **discomorphella** » Sat Aug 17, 2013 9:51 am

Hi Waldo,

Brilliant. I've not used the Olympus product, but your results are certainly equivalent to the best that I've seen using AutoQuant and Huygens. Which objective did you use? And does cellSens allow for superresolution? And did Olympus say which algorithm they use?
Ironically, one of the main coders from AutoQuant is now employed in the CAD group at my company, and it was quite interesting to discuss PSF restoration with one of the principals when I was trying it out in lab. I have not had the opportunity to employ it on any homemade stacks, just no time, but your beautiful fern flourescence shot is certainly motivation to go and grab a stack.
There are some Matlab 3D MLE restoration implementations floating around the web. I have not tried them yet but if you had access to MATLAB its probably considerably less expensive than the Olympus or other dedicated software but will require a lot more programming on your part.

David

pwnell · Post by **pwnell** » Sat Aug 17, 2013 10:42 am

Hey David,

Thanks - I am quite pleased myself. Still not 100% convinced it is necessarily "prettier" than the normal stacking but it most certainly seems more accurate from a biology perspective.

For this photo I used the LUCPLFLN 40x/0.6 for its ability to have a correction between 0 - 2mm. I am unsure whether cellSens does support superresolution. The algorithm is Advanced Maximum Likelihood Estimation in constrained iterative mode. This photo did not have spherical aberration corrections enabled.

soldevilla · Post by **soldevilla** » Sat Aug 17, 2013 12:34 pm

In astronomy we use various softwares that include deconvolution. Some of them are for free, as CADET. Anyway, I have not ever gotten a correct result using deconvolution, I find it very difficult to quantify the defects of an image to reverse it. Instead Deconvolution,I use wavelets. And I use photoshop high pass filter, the "poor man" wavelet filter.

The picture you show also includes an enhancement for false color, very easy to do in photoshop too.

pwnell · Post by **pwnell** » Sat Aug 17, 2013 1:14 pm

soldevilla wrote:The picture you show also includes an enhancement for false color, very easy to do in photoshop too.

Which false colour?

soldevilla · Post by **soldevilla** » Fri Aug 30, 2013 10:18 am

Sorry, I do not get unread messages correctly.

The false color I get with a gradient layer and placing the colors to my liking. This is a powerful tool because it allows just highlight histogram levels we want to make it visible.