GaussianMixture: Split component...

Splits one component of the selected GaussianMixture into two components.

The selected component is split based on a PCA analysis of its covariance matrix. The new means are situated around the old components mean, 1σ apart in the first principal components direction, and the new covariances are constructed with information from the old covariance matrix.

The details of the algorith are described in Zhang et al. (2003).

© djmw, November 22, 2010