A split-and-merge framework for 2D shape summarization

An algorithm for the representation and summarization of a 2D shape is presented. The shape points are modeled by ellipses with very high eccentricity in order to summarize the contour of the shape by the major axes of these elongated ellipses. To this end, at first, a single ellipse is fitted to the shape which is then iteratively split to a large number of highly eccentric ellipses to cover the shape points. Then, a merge process follows in order to combine neighboring ellipses with collinear major axes to reduce the complexity of the model. Experimental results showed that the proposed algorithm provides a shape summary which not only overcomes the representation of a shape by a Gaussian mixture model but also is largely more accurate with respect to the progressive probabilistic Hough transform for shape representation. It must be noted that, for our method, a shape is a unordered set of points describing the contour of a region.