Modelling of Inexact 2-d Shapes Using Real-coded Genetic Algorithms

Inexact Shapes-where a single model is capable of representing a class of shapes-provide an abstraction that is useful in conceptual design, object recognition, morphing, etc. In this work we present an axis-based model for ambiguous shape modeling. This is done by perturbing the links in an axial model of a contour in arbitrary ways. We demonstrate an optimization application in this paper where we start not from a parametrized or relatively static geometry, but from a exibly deened class of shapes. In typical geometric optimization tasks, often the geometry is so far constrained in deening the problem that many better optima actually may lie outside the search space. This paradigm provides a means for capturing a wider class of feasible shapes. A real-coded Genetic Algorithm has been used in this work to implement the visualization and optimization of the inexact shapes. Results are shown for a wide range of contour shapes from biological objects to discrete parts and buildings.