Genetic algorithm with competitive image labelling and least square

摘要

A multi-modal genetic algorithm using a dynamic population concept is introduced. Each image point is assigned a label and for a chromosome to survive, it must have at least one image point with its label. In this way, the genetic algorithm dynamically segments the scene into one or more objects and the background noise. A Repeated Least Square technique is applied to enhance the convergence performance. The integrated algorithm is tested using a 6 degrees of freedom template matching problem, and it is applied to some images that are challenging for genetic algorithm applications.