Interpretation Models and reasoning Strategies in
Scene analysis
Toni Schenk
Department of Geodetic Science and Surveying
The Ohio State University
1958 Neil Ave., Columbus, OH 43210
ABSTRACT
The presentation begins with a brief overview of reasoning strategies, ranging from
simple constructs to more sophisticated inference schemes. This is followed by a
discussion of how reasoning fits in to the vision paradigm. Next, basic assumptions are
discussed, including the fact that input data like digital imagery for solving vision tasks—
or photogrammetric problems for that matter—are incomplete and ambiguous. Therefore,
only inferences can be drawn, not deductions. I also believe that the (automatic) solution
of photogrammetric problems does not permit shortcuts. For example, an interpreted
aerial scene cannot be obtained in one single step from the digital image. Rather, several
processes are involved which result into increasingly more abstract representations.
Inference methods must deal with knowledge and information derived from the input
data. Ultimately, the data driven processes must be supported (guided) by domain
specific knowledge.
The second part of the presentation will focus on abductive inference and its application to
recognizing objects in aerial scenes. After a brief overview and the motivation for using
abduction, specifications of a system under development are discussed. The presentation
ends with conclusions, including a brief discussion of numeric vs. symbolic reasoning.
Schenk - 1