Update the docstring of MAR weight as discussed yesterday:
In multi resolution reconstruction, change the interpolation method from scipy.zoom() to skimage.rescale(), which yields a smoother interpolation with less high-frequency artifacts.
As we observed last week, this results in faster convergence of the 3D shepp Logan problem (highest resolution recon: 18 iterations -> 15 iterations) without sacrificing image quality.
This also reduces the software dependency. The mbircone source code does not require scipy anymore.
This PR contains two components:
Update the docstring of MAR weight as discussed yesterday:
In multi resolution reconstruction, change the interpolation method from scipy.zoom() to skimage.rescale(), which yields a smoother interpolation with less high-frequency artifacts. As we observed last week, this results in faster convergence of the 3D shepp Logan problem (highest resolution recon: 18 iterations -> 15 iterations) without sacrificing image quality. This also reduces the software dependency. The mbircone source code does not require scipy anymore.