registration in starfish, Part 1: zero out wrapped values

[berl]

The lone registration module fourier_shift uses scipy's ndimage.fourier.fourier_shift, which wraps intensity values from one side of the image to the other. I've tried to fix this by tacking on some code to zero out those wrapped values in berl-zero-wrapped-intensities.

[dganguli]

I've also fixed this (not inn a branch yet) by cropping the images. Is zeroing out the values preferred to cropping? Should we support both options?

[berl]

@dganguli isn't the cropping risky once you get into multi FOV data? it changes the shape of the FOV which would/should propagate all the way up to the image stack. E.g. you do the crop, use the result to localize a spot and then want to go back to the unprocessed data and check the raw intensity values in a neighborhood around the spot. This would be trivial with the zero'd result (or in the IntensityTable with #410 ) and annoying with a cropped result.

[dganguli]

that sir, is an extremely good point.

[ttung]

++ zeroing out.

[ttung]

@shanaxel42 isn't this done?

[shanaxel42]

By this are you referring to padding? Because that is not done