OPTIMIZATION CRITERIA BASED ON SCATTER
- Within-class scatter defines the scatter of samples around their
respective means.
- Between-class scatter defines scatter of expected vectors
around the global mean.
- Mixture-class scatter is the overall scatter obtained from
the covariance of all samples:
where
is the overall mean.
- Optimizing criteria used to obtain the transforms is a combination of
within-class scatter, between-class scatter and overall scatter.
- Transformation matrix is given by:
where
are the eigenvectors corresponding to non-zero eigenvalues.