Add deconvolution and texture analysis notebooks for I2k 2023

[elevans]

This PR adds two new jupyter notebooks to the use case section in the PyImageJ documentation. These notebooks do the following:

• 3D Deconvolution
• GLCM/Haralick feature texture analysis

Both of these notebooks only need ImageJ2 and no legacy :tada: so we don't need any changes to the code base. I also reorganized the use case section into sub sections:

• imagej-ops
• Integration
• Segmentation
• Other use cases

The goal is to use these notebooks in the upcoming I2K 2023 workshop.

[codecov[bot]]

Codecov Report

All modified lines are covered by tests :white_check_mark:

Comparison is base (f8c7909) 77.65% compared to head (de1ba3a) 77.65%.

:exclamation: Current head de1ba3a differs from pull request most recent head b5e16e2. Consider uploading reports for the commit b5e16e2 to get more accurate results

Additional details and impacted files

```diff @@ Coverage Diff @@ ## main #290 +/- ## ======================================= Coverage 77.65% 77.65% ======================================= Files 16 16 Lines 1947 1947 ======================================= Hits 1512 1512 Misses 435 435 ```

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

[gselzer]

I have a couple small suggestions, just about organization. Take or leave either of them: 1) In the Decon notebook, I think it would be really nice to display the original and deconvolved images side-by-side, so you can compare the difference more easily. 2) In the GLCM notebook, I'd suggest displaying the image right away - you talk about the image but unless I can immediately relate that to what I see in the image I immediately forget the details you listed off.

[bnorthan]

Hi @elevans

I would change the paragraph that says "this notebook utilizes the Richardson-Lucy Total Variation (RLTV) algorithm, which has significantly improved axial resolution over the standard RL algorithm"

The RLTV algorithm is used to limit noise, which can be a problem with the standard RL algorithm, however there is not alarge difference in 'resolution' between the standard and TV regularized version (as a side note resolution is a pretty nuanced topic and technically deconvolution improves 'contrast based' resolution methods, like the Rayleigh criteria, but does not restore out of band frequencies to a significant extent).

See below description and figure from my deconvolution/deep learning workshop. Even though the workshop used my opencl python version of RL, the figure linked below was actually generated a long time ago, for the 2015 ImageJ Conference, and used the version of RL from imagej-ops.

https://true-north-intelligent-algorithms.github.io/deconvolution-gpu-dl-course/20_deconvolution/0_intro_to_decon.html#richardson-lucy-with-total-variation-regularization-noisy-image

[elevans]

Thanks @bnorthan! I'll update my text to state that RTLV limits the noise that standard RL can amplify.

[hinerm]

Awesome stuff @elevans .. you taught me some stuff today! 😄