Boyer -1
1
The Role of Perceptual Organization
in
Automated Scene Analysis
Kim L. Boyer
Signal Analysis and Machine Perception Laboratory
Department of Electrical Engineering
The Ohio State University, Columbus, Ohio, U.S.A.
Email: kim@ee.eng.ohio-state.edu
Abstract
In this brief position paper, we highlight the role of perceptual organization in
automated scene analysis and argue that it is, indeed, the sophistication of a vision
system’s perceptual organization capability that underlies the sophistication and
performance of the system’s performance as a whole. No other single aspect of the
system design is apt to be so critical to its success. We will address these issues
by considering the issues raised by-the Working Group Chairmen. Thus, this paper
will dispense with the traditional organizational structure and simply consider the
questions 1 . The questions as phrased by the Chairmen appear in italics , while our
responses appear in plain type. More detail regarding this work can be found in the
list of related publications given at the end of the paper.
The Questions
• Matching and grouping are fundamental processes in many tasks in computer
vision, sensor orientation, and scene registration. In sophisticated analysis
systems working with imagery of realistic size, matching and grouping of prim
itives often dominates processing time. Are there any new ideas, particularly
from photogrammetry, to improve the combinatorics?
We cannot speak for the photogrammetrists, but we do submit that
the voting and Gestalt graph methods accompanied by the Bayesian
Perceptual Inference Networks (PINs)as described in our prior work
and highlighted in the talk offer considerable relief from the combi
natorial burden. For instance, typical execution times range from 7
1- \Ve respond only to those questions in which we have reasonable expertise.