Hierarchical EGI: a new method for object representation

This paper proposes a new object representation scheme called hierarchical extended Gaussian image (HEGI). The HEGI can be used for describing nonconvex as well as convex objects. The HEGI description can be constructed as a tree where each leaf node corresponds to an extended Gaussian image (EGI) description of a convex part resulting from the recursive convex hull decomposition of an object. An object's HEGI consists of three major components. The first component is a Gaussian image set representing the object's parts. The second component is a relational set keeping the intersecting curves between the different parts. The third component is an attributed graph storing the part attributes and interpart relational attributes. The HEGI is a hierarchical and intrinsic scheme with decreasing spatial resolution at each higher layer.