This releases fixed important bugs in convolutions and add normalized moments computations. Full ChangeLog: * Better error checking * Fix interpolation of integer images using order 1 * Add resize_to & resize_rgb_to * Add coveralls coverage * Fix SLIC superpixels connectivity * Add remove_regions_where function * Fix hard crash in convolution * Fix axis handling in convolve1d * Add normalization to moments calculation
In debug mode, this performs checks for memory access. In release mode, it is optimized away to raw pointers.
This achieves robustness against image scalings.
Now, for small images, the general purpose code is used, which checks performs a lot more bounds checking and does not access memory outside the image region.
This is another advantage of C++: many small wrappers check bounds in debug case, optimize away in release mode.
The axis argument was wrongly handled in convolve1d. In the 2D case, the axes were switched, in higher dimensions, the mistake was more complex.
Arguments were not checked in Python, triggering a segmentation fault later.
For consistency with numpy, output arguments are now called `out`.
python3-numpy installs but later triggers an error This currently does not install matplotlib on Python 3. Thus, those tests will be skipped.
"pip install scipy" was timing out
There were tests which were not getting executed
The previous method was still not enough.
- Rename test with name clash. - Test cdilate early exit - Tag legacy compatibility code as no cover
10x faster (even if this function is generally called for small radii).
Coverage picked up that the sobel test was not really meaningful and some lines in shape.py were not being covered.