Legacy Isotropic Whitening

[garrettwrong]

This PR implements the legacy pre-whitening code from MATLAB.

Also includes two other cleanups. First, some of the Filter methods (_create_filter) are simplified/reconciled and enforced with abc. Second, the ImageAccessor image count attribute is changed from num_images to n_images. This allows an ImageAccessor to be passed in place of an Image stack in the static methods (epsdR, epsdS) I've ported in this PR, but is otherwise mostly an internal change.

[garrettwrong]

Cleanup and changed some random camel casing towards our more typical var names (bgRadius->bg_radius, batchSize->batch_size).

Need to finish the GPU interop and begin testing.

[codecov]

Codecov Report

Attention: Patch coverage is 80.37736% with 52 lines in your changes missing coverage. Please review.

Project coverage is 90.43%. Comparing base (bb67480) to head (c48d1a4).
Report is 37 commits behind head on develop.

Files with missing lines	Patch %	Lines
src/aspire/noise/noise.py	89.36%	15 Missing ⚠️
src/aspire/numeric/mkl_fft.py	0.00%	15 Missing ⚠️
src/aspire/numeric/base_fft.py	50.00%	7 Missing ⚠️
src/aspire/numeric/pyfftw_fft.py	50.00%	5 Missing ⚠️
src/aspire/source/image.py	79.16%	5 Missing ⚠️
src/aspire/image/image.py	89.47%	4 Missing ⚠️
src/aspire/commands/cov3d.py	0.00%	1 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #1223      +/-   ##
===========================================
- Coverage    90.63%   90.43%   -0.21%     
===========================================
  Files          132      132              
  Lines        13705    13939     +234     
===========================================
+ Hits         12422    12606     +184     
- Misses        1283     1333      +50     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[garrettwrong]

Initial tests of legacy_whiten reproduce MATLAB pretty closely for a small sample of real data at 129 and 128 pixels. Notebook including minimal scripts for both repos is attached.

CompareWhiten.pdf

I will start some larger runs in the background.

This just needs some self review and should be ready for further consideration.

[j-c-c]

This looks great. Just a couple small things.

[j-c-c]

Looks Good!

[garrettwrong]

During my optimization work I may have spotted a small bug in the python corr accumulation loop that duplicates accumlations (but the same bug applies to the denominator count accumulation, so I think it more/less cancels).(fixed) Once I finish testing the optimized code I might merge it into here... (this afternoon maybe).

[janden]

Nice work! Some questions and remarks (mostly naming) here.

[garrettwrong]

I think I addressed the suggestions. This also makes the normalization optional.

I will begin retesting this against MATLAB and with the full size recon.

[garrettwrong]

Reproduced the test case in the notebook. Experimental data recon will finish tomorrow, but I don't anticipate any surprises given the test batches were good.

[garrettwrong]

Experimental recon was also good.

[janden]

Great! Just one name suggestion and we should be good to go.