Combined 3D-2D object description and representation

In this paper we present a representation form for 3D objects by a number of different 2D views. Experiments were carried out to identify viewing positions for relevant 2D-views and number of views necessary for certain retrieval results. This approach allows similarity-based retrieval between 3D objects and 2D-images in the context of MPEG-7 description. Within the representation, basic 2D Descriptors can be combined into a Multiple View Descriptor and used to characterize a 3D-object. Using the Multiple View Descriptor, it is also possible to match 2D views taken from unknown viewing position with a representation of a 3D object. Furthermore, beyond the MPEG-7 related applications, we present a method to obtain a partial 3D reconstruction of the original object if it is only represented by depth maps relating to a number of 2D views.