A Comparison of Parametric Contour Spaces for Interactive Genetic Algorithms

Technical Report: OSU-ACCAD-6/01-TR1 (429k pdf)

Matthew Lewis, The Advanced Computing Center for the Arts and Design, The Ohio State University

Richard Parent, Department of Computer and Information Science, The Ohio State University

Abstract

A common criticism of many visual interactive evolutionary design systems is that most of what they generate "all looks the same". For any given system's output, formal visual characteristics can frequently be identified that are shared by a majority of designs produced by the system. These visual traits can either be expressed in terms of shared design features (e.g., all corners are sharp) or a near complete absence of traits (e.g., no curved edges on any individuals). Several simple shape representations are presented here to illustrate and explore the relationship between the use of various approaches for constructing design spaces and the visual traits which result. It is argued that the relative signature of a system is determined by the amount that the design space representation biases both the possibility and probability of the emergence of specific visual design attributes.

Reflected Contour Shape Movies

(quicktimes, around 700k)
Five vertices, horizontally displaced
Twenty vertices, horizontally displaced
Five vertices, horizontally and vertically displaced
Twenty vertices, horizontally and vertically displaced
3-20 vertices, horizontally and vertically displaced
One hundred vertices, offset horizontally with noise

Polar Contour Shape Movies

(quicktimes, around 700k)
Eight vertices, distance displacement
Eight vertices, distance and angle displacement
Thirty-two vertices, distance displacement
Thirty-two vertices, distance and angle displacement
4-32 vertices, distance and angle displacement
One hundred vertices, distance displacement with noise

ACCAD