Avoid empty kernel

[mawc2019]

In this PR, the situation of an empty kernel is avoided and #23 is fixed. The variable shape is renamed as kernel_size to avoid confusion with another variable kernel_shape and the Python command .shape.

[stevengj]

What about non-square kernels, e.g. kernel_size[0] == 0 and kernel_size[1] == 3?

[mawc2019]

This PR did not take care of this situation. The fix should be revised as something like:

  if kernel_size in np.array([[2,2], [2,1], [1,2], [2,1]]):
    return np.ones(kernel_size, dtype=np.uint8)

  if kernel_size[0] == 0 or kernel_size[1] == 0:
    return np.array([[1]], dtype=np.uint8)

[mawc2019]

In the revised PR, I used the following lines instead:

  if kernel_size[0] == 0 or kernel_size[1] == 0:
    return np.array([[1]], dtype=np.uint8)
  elif kernel_size in np.array([[2,2], [2,1], [1,2], [2,1]]):
    return np.ones(kernel_size, dtype=np.uint8)
  else:
    rounded_size = np.round(diameter / pixel_size - 1) * pixel_size

[stevengj]

Do we actually support non-square kernels?

[mawc2019]

Yes, the code including its early versions supports non-square kernels, which typically arise if a pixel has different sizes in x and y directions.

[stevengj]

Closed by #30