Swarm Methods of Image Segmentation

Abstract

A distinctive feature of the thematic image segmentation of onboard optoelectronic surveillance systems is the search for rational solutions in the multidimensional space of alternatives. Searching for rational solutions in a multidimensional space of alternative is peculiarized by non-linearity, non-differentiation, multi-extremality, ravine surface, lack of the analytic expression of objective functions, high computational complexity, high dimensionality of the search space, and the complex topology of the region of acceptability. Finding a solution by exact methods of optimization is complicated by non-linearity, non-differentiation and lack of the analytic expression of the objective function. High computational complexity, high dimensionality of the search space, and the complex topology of the region of acceptability lead to wasting valuable time when using precise optimization methods. Currently, methods of searching for global extremum are being developed, which provide the convergence to the exact solution of the optimization problem, provide the optimal (minimum or maximum) value of the objective function. Such methods include meta-heuristic optimization methods, which, unlike classical optimization methods, can be used even if there is no information about the nature and properties of the objective function. Meta-heuristic methods have the following features: they manage the process of finding the optimal solution; they efficiently study the search space to find the optimal solution; they use simple local search procedures and complex training processes; they are approximate methods and, as a rule, non-deterministic; they take into account the probability of trapping in a limited search space; they are universal (solve various application problems); they use a priori information to find the optimal solution. Let us detail the use of the variety of meta-heuristic optimization methods (ant colony optimization (ACO) and artificial bee colony optimization (ABC) for the thematic image segmentation of onboard optical and electronic surveillance systems.