CDI: find method to estimate masked (mostly high-angle) pixels and avoid diverging estimates for high-angle intensity
Diverging intensities at high angle can create checkerboard-type patterns in the object, and they are difficult to remove because they are not removed by the support limits (except maybe if the support is exactly tight).
@chushkin it would be interesting to see if the GPS algorithm, which includes a smoothing parameter in object space (also in Fourier space) can be used to remove some of the checkerboard effects due to high angle pixels(smoothing in real space is equivalent to multiplying by a window function in Fourier space)
Alternatively, maybe the approach used for gaps would also be relevant:
- estimate the values expected for all pixels by extrapolation (should be possible at high angle if there are no sharp facets)
- use the projections algorithm with a large confidence interval (in that case maybe from 0%)