Clustering by Maximizing the Difference between Feature Covariance?

Is there any clustering algorithm that attempts to find groups of samples that have maximally different feature covariances? Want to make sure I'm not reinventing the wheel here (or doing something really dumb)