If we have a label image:
[[1, 2],
[3, 4]]
Then the footprint:
[[0, 0, 0],
[0, 1, 1],
[0, 1, 1]]
will not discover the edge (2, 3), even though that edge should be
present when `connectivity=2`.
This commit fixes the bug by using a complete footprint, without the
top/left surfaces zeroed out.
See also:
https://github.com/scikit-image/scikit-image/pull/1826#issuecomment-164597442
When using generic_filter to build a graph, the numerical filter
output is actually ignored. Building a full array, even of the
smallest dtype, was wasteful. Using stride_tricks solves this by
creating a single-element buffer into which all output is fed.
Fix some sphinx warnings
Add documentation build to test
Add documentation build to test
Remove change in numpydoc
Remove change in apigen
Add makefile target for html and add to travis script
Add a makefile target for html and add to travis script
Fix more sphinx warnings
Vighnesh Birodkar