Particle Swarm Optimization and Image Analysis

Particle Swarm Optimization (PSO) is a simple but powerful optimization algorithm, introduced by Kennedy and Eberhart (Kennedy 1995). Its search for function optima is inspired by the behavior of flocks of birds looking for food. Similarly to birds, a set (swarm) of agents (particles) fly over the search space, which is coincident with the function domain, looking for the points where the function value is maximum (or minimum). In doing so, each particle’s motion obeys two very simple difference equations which describe the particle’s position and velocity update. A particle’s motion has a strong random component (exploration) and is mostly independent from the others’; in fact, the only piece of information which is shared among all members of the swarm, or of a large neighborhood of each particle, is the point where the best value for the function has been found so far. Therefore, the search behavior of the swarm can be defined as emergent, since no particle is specifically programmed to achieve the final collective behavior or to play a specific role within the swarm, but just to perform a much simpler local task. This chapter introduces the basics of the algorithm and describes the main features which make it particularly efficient in solving a large number of problems, with particular regard to image analysis and to the modifications that must be applied to the basic algorithm, in order to exploit its most attractive features in a domain which is different from function optimization.