Segmentation with Superquadrics

A common underlying task of most recognition applications is building the scene description in terms of symbolic entities. A challenging problem in scene understanding is segmentation, where each piece of information must be mapped either to a shape primitive or discarded as noise. At the same time, there should be a minimum number of such primitives applied, so as to get as compact a description as possible. The absence of the domain knowledge further makes it more difficult, as ambiguities arise due to multiple representations and incomplete data.