Inertia Tensor in real units may not be sorted!

[jlegrand62]

When computing inertia tensors, eigenvalues are returned sorted in decreasing order by SVD.
But if we ask for real values, order may change due to non-homogeneous voxelsize!

In spatial_image_analysis.py, change:

coord = coordinates_centering3D(xyz, center)
# compute the variance-covariance matrix (1/N*P.P^T):
cov = compute_covariance_matrix(coord)
# Find the eigen values and vectors.
eig_val, eig_vec = eigen_values_vectors(cov)
# convert to real-world units if asked:
if real:
    for i in xrange(3):
        eig_val[i] *= np.linalg.norm( np.multiply(eig_vec[i],self._voxelsize) )

into

coord = coordinates_centering3D(xyz, center)
# convert to real-world units if asked:
if real:
    coord = np.multiply(coord, self._voxelsize) )
# compute the variance-covariance matrix (1/N*P.P^T):
cov = compute_covariance_matrix(coord)
# Find the eigen values and vectors.
eig_val, eig_vec = eigen_values_vectors(cov)

to be sure to keep the eigenvector sorted by eigenvalues !