Generative noise models (test cases) used in the randomized registration trials of Experiment 6.

This table defines the covariances used to generate the zero-mean, multi-variate, Gaussian noise applied to the source shape in each test case of Experiment 6. The table defines both the eigenvalues (λ1, λ2, λ3) (magnitude) and the eigenvectors (orientation) of the covariance matrices for each test case. Orientation “Surface” defines the eigenvectors relative to the orientation of the surface at each source point, where λ1 is the variance in the surface-normal direction and (λ2, λ3) are the variances in the surface-parallel directions. Orientation “Random-Global” defines the matrix of eigenvectors to be a randomly generated rotation matrix, which is applied globally over the source shape (i.e. every point in the source shape has the same noise model). Orientation “Random-Per-Point” is similar to the “Random-Global” case except that every point in the source shape is associated with a different randomly generated rotation (i.e. every point in the source shape has a different noise model). Orientation “-” means that the noise model is isotropic and thus unaffected by the choice of eigenvectors.
Generative noise models (test cases) used in the randomized registration trials of Experiment 6.