Smarter transforms

[jzaremba]

At least for in vivo imaging data, transforms calculated between sessions are intended to correct for two things:

1. Movement of the imaged field of view from session-to-session (This should just be rotation/translation)
2. Warping of the tissue itself over time. (This should allow rotation/translation/shear?).

Two issues with the affine transforms we apply now:

1. There is no biological reason the scale factor between sessions should be anything other than 1.
2. Any kind of shear would be do to changes in the tissue as a whole, not the cells themselves, so it should effectively only be applied to the spacing between ROIs, not the shape of the ROIs themselves.

Ideally, calculate/apply transforms separately on the ROI centroids and the ROI shapes. At least only apply the calculated shear to the centroids.