Robust change-point detection and segmentation in data streams

We address the problem of robust detection of those points in time where the statistical properties of an observed time series change. This is a critical theoretical problem in modern statistical signal processing with significant applications in multiple access communication systems and segmentation of multimedia signals, as well as in hidden signal detection, fault detection and diagnosis. We develop new, robust algorithms which maintain a performance close to the corresponding optimum (maximum likelihood) algorithm in the entire class of symmetric, alpha-stable signals and noise. Our algorithms are illustrated via computer simulation, while application on real radar sea-clutter data corroborates our findings