Body-centered representation and perception
Intuitive notions of shape, object, and part are informative for perception and modeling. Intuitive sense of shape is a volume concept, not a surface concept; shape relates surfaces typically as ratios of diameters.
We propose a natural 3-d definition based on function. Parts are volume elements derived as a visual observable from a fundamental part definition, based on C1 continuity of generalized cylinder (GC) primitives. Parts observable from quasi-invariants from GC primitives implement a useful object-background discrimination.
The value of such a shape representation is that the concepts of the representation map one-one into nodes in a Bayesian network for perception.
Unable to display preview. Download preview PDF.
- 1.T.O. Binford. Survey of Model-Based Image Analysis Systems. Robotics Research, Vol. 1, No. 1, March 1982.Google Scholar
- 2.T.O. Binford, Tod S. Levitt, W.B.Mann. Bayesian Inference in Model-Based Machine Vision. Proc Workshop on Uncertainty in AI, AAAI87, 1987.Google Scholar
- 3.T.O. Binford. Shape. Encyclopedia of Artificial Intelligence, Second Edition, 1992, Wiley.Google Scholar
- 4.D.D. Hoffman and W. Richards. Representing smooth plane curves for recognition. Proc AAAI, 1982.Google Scholar
- 5.J. Mundy. Object Recognition: The Search for Representation. In these proceedings.Google Scholar
- 6.H. Sato and T.O. Binford. Finding and Recovering SHGC Objects in an Edge Image. CVGIP: Image Understanding; V 57, pp 346–358, 1993.Google Scholar