Non-convex onion-peeling using a shape hull algorithm

The convex onion-peeling of a set of points is the organization of these points into a sequence of interpolating convex polygons. This method is adequate to detect the shape of the "center" of a set of points when this shape is convex. However it reveals inadequate to detect non-convex shapes. Alternatively, we propose an extension of the convex onion-peeling method. It consists in representing a set of points with a sequence of non-convex polylines which are computed using the A-shape descriptor. This method is applied to robust statistical estimation. It is shown that it makes the estimators robust to the presence of outliers by removing suspect samples from the available population. (C) 2004 Elsevier B.V. All rights reserved.