Drawbacks of the CloughTocher2d Interpolator from Scipy

[jruebsam]

The CloughTocher Interpolation in the replace_outliers function in filters.py
does not give the same results when the image is rotated on both axis
see for example this script:
ctttest.zip

Furthermore NaNs at the boundary are replaced by the fill_value 0
which is not optimal but the fastest working solution so far.

Another interpolation scheme should be used,
in order to avoid these drawbacks.

[jruebsam]

Update: This seems to be a generel problem of scipy.interpolate.griddata functions.. .

possible solutions .. .
http://stackoverflow.com/questions/21690608/numpy-inpaint-nans-interpolate-and-extrapolate
http://scikit-image.org/docs/dev/auto_examples/filters/plot_inpaint.html