Feature extraction from the Hermitian manifold for Brain-Computer Interfaces

Riemannian geometry-based methods have shown to be effective in many sorts of Brain-Computer Interface (BCI) applications, but are only capable of measuring the power of the measured signal. This paper proposes a set of novel features derived via the Hilbert transform and applies them to the generalized Riemannian manifold, the Hermitian manifold, to see whether the classification accuracy benefits from this treatment. To validate these features, we benchmark them with the Mother of All BCI Benchmarks framework, a recently introduced tool to make BCI methods research more reproducible. The results indicate that in some settings the analytic covariance matrix can improve BCI performance.