Adaptive M-estimators for robust covariance estimation

Abstract

Robust covariance estimates are required in many applications. Here, a promising adaptive robust scale estimator is extended to this problem and compared to other robust estimators. Often the performance analysis of covariance estimators is performed from the perspective of the final application. However, different applications have different requirements, hence we make a comparison based on some general metrics. Results show that the adaptive scheme shows good per- formance under the nominal case and graceful degradation in performance with increasing levels of contamination. robust estimation, covariance, M-estimators.