On the development of a fuzzy genetic framework for the segmentation of synthetic aperture radar images.

摘要

This thesis concerns the development of an image segmentation framework based on fuzzy set optimization under the governance of a genetic algorithm. Fuzzy segments are systematically evolved and reconfigured to minimize a cost functional, subject to constraints that maintain topological correctness of the fuzzy sets. Segment geometry is conditioned using Sethian-Osher curvature flow shaping constraints, which operate on the level sets of the fuzzy membership function. Cell populations are divided and recombined under the direction of a global cost optimizer running concurrently with algorithms responsible for cellular mutation. A discrete lattice structure is used to enforce a broader set of constraints on segment geometry and fuzzy set topology. A Java(TM)-based software implementation of the segmenter is used to demonstrate the power and utility of the segmenter, and to compare the framework with existing likelihood-based SAR segmentation algorithms to validate claims about the superiority of the new approach.