Shape is an important cue in rcognizing objects in pictures. Shape is, however, difficult to describe and use in computer programs for object recognition, because objects appear different in images taken from different viewpoints. This effect is particularly pronounced for curved three-dimensional objects. Recent work has shown that, for some surfaces, it is possible to compute descriptions that draw only on information from a single picture, and are the same for every possible view of a given surface. These invariant descriptors instantly identify an object, up to a known ambiguity. This research investigates advance systems for object recognition that use a combination of invariant descriptors and color information to recognize curved three- dimensional objects, drawn from a large number of known examples, in pictures taken from an unknown viewing position.//
