Missed that in the notes, thanks for the notice. Looks like I need to update. If/when we consider moving to Numpy 1.7.x as a dependency instead of 1.6.x, I'll try to remember to bring this up again. On Thursday, April 18, 2013 7:52:06 PM UTC-5, Christoph Gohlke wrote:
On 4/18/2013 5:41 PM, Josh Warner wrote:
It appears there are several scikit-image functions which have need of padding input. As far as I can see, these generally do so internally, with functions or routines that aren't exposed. From a quick search, the following functions in scikit-image all pad using their own internal methods:
|feature.match_template| : mean value |filter.lpi_filter| : asymmetric zeros |morphology.greyreconstruct| : symmetric zeros or constant value |morphology.watershed| : |pad| child function for symmetric zeros |transform.radon| : symmetric zeros
More broadly, there doesn't appear to be a publicly exposed API for padding NumPy image arrays in any packages I've come across. |scipy.ndimage| clearly has this functionality, via the common |mode| parameters, but it isn't exposed for general use.
I've put some effort into making a pure Python 2d / 3d capable padding function for personal use, and most of the |scipy.ndimage| modes are available. Preliminary gist here wouldn't be hard to wrap into a PR (likely not optimal, but functional): https://gist.github.com/JDWarner/a26f77dae983fa1faba9
Would such functionality be desirable?
Numpy has a pad function since version 1.7. <http://docs.scipy.org/doc/numpy-dev/reference/generated/numpy.pad.html>
Christoph