Performance of real-world applications in computer vision and image understanding largely depend on the efficiency of low level tasks involved. One of such low level tasks is the image segmentation process, which is from the combinatorial optimization problem domain. The survey paper emphasizes the recent use and increasing growth of the metaheuristics approach based on natural computing inspired by natural and biological systems for solving the image segmentation specific to various domain-specific applications. The findings of recent studies of such approximate algorithmic approach, largely based on heuristics, to obtain near-optimal solutions are showing strong evidence to the fact that they can be applied to almost all approaches of various image segmentation techniques, such as edge detection, region-oriented, morphological watersheds and clustering. The paper works towards bridging the gap with regard to drawbacks of the state-of-the-art solution of the various segmentation techniques and the available metaheuristic techniques and technologies like particle swarm optimization, Ant Colony Optimization (ACO) and Genetic Algorithms (GA). Finally, it proposes modeling such automated system for learning algorithms based on signals-to-symbol model.

Researchers are aiming to automate very complex problems using computers, which can be made possible only by very time-efficient algorithms. The exact deterministic algorithms may take ages to solve combinatorial optimization problems like the famous Traveling Salesman Problem (TSP) for larger search space, which have been proved to be of class NP. Class NP consists of all those problems whose solutions can be found in polynomial time on a non-deterministic Turing machine. Since such a machine is not practically feasible, it means that an exponential algorithm can be written for an NP-problem; also, there is no certainty as to whether a polynomial algorithm exists or not [1]. Computing optimal solutions for many combinatorial optimization problems are intractable, and known as NP-hard. Usually, we are pleased with good solutions rather than having no feasible solutions at all. It has been time and again proved in recent studies that heuristic and metaheuristic algorithms, which suggest some approximations to the solution of optimization problems, are as good a solution as possible for the problem instances. There are two important concepts in metaheuristics known as intensification and diversification, which largely determine its behavior. They are in some way contrary and complementary to each other.

The paper [2] by Christian Blum and Andrea Roli introduced a framework, I&D frame, to put different intensification and diversification components into relation with each other. Some classic NP problems are TSP and Quadratic Assignment Problem (QAP), which have been successfully solved using metaheuristics. Interested readers can refer to paper [3], which solves TSP using Ant Colony Optimization (ACO) metaheuristics. One of the most successful types of artificial intelligence techniques known as swarm intelligence [4, 5, 6] solves QAP using hybrid ant colony system and Genetic Algorithm (GA) [7]. Research on ACO algorithms has led to a number of other successful applications to combinatorial optimization problems; the results of which indicate that it is possible to arrive at high quality solutions in reasonable time. In addition to single solution search algorithms such as descent local search, greedy heuristic, simulated annealing and tabu search, there is a growing interest in population-based metaheuristics well inspired by the behaviors of natural systems. These metaheuristics include evolutionary algorithms, evolution strategies, genetic programming, ant colonies, scatter search and particle swarm optimization [8].

Science and Technology Journal, Metaheuristics Optimization, Image Segmentation Techniques, Ant Colony Optimization, ACO, Quadratic Assignment Problem, QAP,Traveling Salesman Problem, TSP, Genetic Algorithms, GA, Synthetic Aperture Radar, SAR.