A MULTI-AGENT APPROACH FOR EDGE DETECTION USING A GENETIC ALGORITHM FOR PARAMETERS’ SPACE EXPLORATION

Abstract

In this paper, an agent based approach for edge detection is presented. It uses the blackboard system as a means of communication between agents. A population of agents is deployed on a two-dimensional representation of an image. Every agent is able to decide whether the pixel on which it is situated belongs or does not belong to the homogeneous region looked for, and thus to exhibit a reactive behaviour: breeding and labelling, or diffusion, allowing the emergence of a complex phenomenon at the global level. This phenomenon is the segmentation of the image. The behaviour of agents is inspired from the natural diffusion phenomenon. The approach has been implemented with the Netlogo platform which is a very powerful agent based simulator, and since the parameters space is very huge, a genetic algorithm has been used to lessen the complexity of the problem.