Computation of "Paganin margin"
The phase retrieval step takes place in the "radios" domain.
If only a few line of each radios are loaded (i.e "small delta_z
"), the filtering step will be inaccurate. Therefore, we have to load enough radios lines to do a proper phase retrieval.
In PyHST, there was a PAGANIN_MARGE
parameter: even if only one slice is reconstructed (delta_z = 1
), data load and phase retrieval would take place on delta_z = 1 + 2*PAGANIN_MARGE
lines of each radio.
I believe this "margin" can be computed automatically for filter-based phase retrieval (ex. Paganin). The filter depends on the following parameters: (energy, distance, delta/beta, pixel size, image dimensions). Inspecting the filter decay will give a clue on how much data we need to load.
Edited by Pierre Paleo