Add the Guided filter to ndimage/filters.py

[lobpcg]

Is your feature request related to a problem? Please describe.
(Fast) Guided filter (https://github.com/topics/guidedfilter) is an edge-preserving smoothing filter like the bilateral filter. It is straightforward to implement and has linear complexity independent of the kernel size. It's included in https://docs.opencv.org/master/javadoc/org/opencv/ximgproc/GuidedFilter.html and https://www.mathworks.com/help/images/ref/imguidedfilter.html
and coded in Python for 2d np arrays in
https://github.com/pfchai/GuidedFilter/
and for PyTorch and Tensorflow for 2d data tensors in
https://github.com/wuhuikai/DeepGuidedFilter

Describe the solution you'd like
Add the (Fast) Guided filter to ndimage/filters.py applicable to np arrays with any dimension >0 in a plurality of directions. The guided filter is based on the box filter, which technically is available as https://docs.scipy.org/doc/scipy/reference/generated/scipy.ndimage.convolve1d.html in mode ‘constant’ if I understand correctly, but could probably more efficiently coded using cumsum, as in https://github.com/pfchai/GuidedFilter/blob/master/guidedfilter.py

Describe alternatives you've considered

• https://github.com/pfchai/GuidedFilter/blob/master/guidedfilter.py is purely Pythonic, but handles only 2d arrays.

• https://github.com/Sundrops/fast-guided-filter is for OpenCV.