Representations of 3D Objects that Incorporate Surface Markings

In many cases, the geometric representation that a recognition system could recover is insufficient to identify objects. When object geometry is simple, it is not particularly distinctive; however, a rich representation can be obtained by mapping the surface markings of the object onto the geometry recovered. If edges are mapped, a representation that is relatively insensitive to the details of lighting can be recovered. Mapping grey levels or color values leads to a highly realistic graphical representation, which can be used for rendering. The idea is demonstrated using extruded surfaces, which consist of a section of a general cone cut by two planes. Such surfaces possess a simple geometry, yet are widespread in the real world. The geometry of an extruded surface is simple, and can easily be recovered from a single uncalibrated image. We show examples based on images of real scenes.