improve background correction

[mattersoflight]

I am reviving this old issue on background correction so we can follow up while working on 1.0.0 release.

Some key principles have emerged in our discussion of background correction strategies:

• Stokes parameters should be normalized per pixel.
• The background needs to be denoised, because the noise propagates to all data. We can denoise background by averaging, or by wavelet denoising.
• In my discussion with @talonchandler, we realized that with the electronic control over input polarization state, one could disambiguate the background introduced by the components between the LC & sample and components between the sample & analyzer. This can be thought of as a variant of using a de Senarmont compensator for calibration. It is worth writing down the model.

We also need to update how estimated background is computed and used to minimize the artifacts such as removal of smooth spatial frequencies of the specimen (see discussion in PR #139).

[ieivanov]

When calculating background from sample images we should be able to select which parts of the data to use

• e.g. average first 10 timepoints
• e.g. first z stack
• e.g. average of z stack This may require that the background dataset is computed by the user before running reconstruction. It may be hard to allow for all option through a config file.

We should be able to specify how the background is applied

• apply same background to all timepoints
• apply same background to all z slices
• apply per-slice background correction of z slices
• apply time-varying background correction? The dimensions of the background stack can dictate how it is applied. E.g. background with XY dimensions is applied to all images; background with ZXY dimensions is applied per-slice across all timepoints; background with T1XY dimensions is applied on all z-slices per timepoint.

[ziw-liu]

Is this fixed as of release 0.2.0?

[talonchandler]

@ziw-liu not entirely...there are still many good ideas in this thread, so I'll keep it open for now.