The Study Of The Particle Swarm Optimization And Its Application In Image Segmentation

Particle swarm optimization (PSO) is an evolutionary computation technique developed by Dr. Eberhart and Dr. Kennedy in 1995, inspired by social behavior of bird flocking or fish schooling. Recently, PSO algorithm has been gradually attracted more attention over another intelligent algorithm. PSO is simple in concept, few in parameters, and easy in implementation. It was proved to be an efficient method to solve optimization problems, and has successfully been applied in the area of function optimization, neural network training and fuzzy control systems, etc. However, both theory and application of PSO are still far from mature.The paper gives a comprehensive study on PSO from the aspect of algorithm mechanism, algorithm modification and its application. Furthermore, image segmentation is the first and foremost problem in image analyzing and mode recognition, and is also a typical stumbling block in image processing. In order to raise its speed, we combined the method of PSO and image segmentation algorithm on valves and therefore proposed several segmentation algorithms based on improved PSO. As we achieve an effective segmentation, we also raised the speed of the parallel searching system. The main content is as follows:(1) The paper surveys PSO algorithm and its basic theories (Optimization method and Evolutionary Computation, EC). First we summarize the generation and development of Optimization method in detail, and emphasize the basic idea, research field and applications. And then we expatiate the emergence, definition and research field, and some typical EC methods, e.g. Genetic Algorithm, Evolutionary Strategy, Differential Algorithm are introduced. At last we introduce PSO algorithm, including its original edition and standard edition, summarize its theoretical and applied research. Monte Carlo method is presented to investigate the ability of particles. The results reveal why the PSO has relative poor global searching ability in the last stage of iteration, it also gives the way how to improve the convergence rate of PSO. Furthermore, nine benchmark functions are used to test the performance of PSO and other popular EC algorithms. The results show that the merits of PSO in terms of the fast convergence rate.(2) In spite of PSO has comparable or even superior search performance for many hard optimization problems with faster and more stable convergence rates, but it can't guarantee to find the global optima in the search space. So the Quantum-behaved PSO (QPSO) algorithm which has power global searching ability than PSO is introduced for improving in this paper. But for QPSO updating the position of particle as whole-item which likes PSO, it also has the problem of the curse of dimensionality. Hence two new hybrid QPSO algorithms with cooperative method (CQPSO and ICQPO) is proposed in this paper for solving this problem. The cooperative method is specifically employed to conquer the"curse of dimensionality", by splitting a particle with composite high-dimensional into several one-dimensional sub-parts. Nine benchmark functions and Maximization of the measure of separability on the basis of between-class variance method (often called OTSU method), a popular thresholding technique, is employed to evaluate the performance of the proposed method. The experiment results show that, compared with the exiting EC methods, the cooperative method helps the new PSO algorithm to get more effective and efficient results. It also conquers the curse of dimension of traditional OTSU method.(3) Based on analysis of the global searching ability of PSO, a new global Gaussian PSO (GGPSO) is proposed to overcome the problem of the premature and low precision of the standard PSO. In this algorithm, combining with global and local mutating method finds an excellent balance between global searching and local searching, which is also guaranteed to converge to the global optimization solution with probability one. Experiment simulations show that the proposed algorithm can not only avoid premature effectively but also has powerful optimizing ability, good stability and higher optimizing precision. For solving image segmentation which is the great importance in the field of image processing, we use Kapur function as the optimization object, and the experiments show that the GGPSO algorithm outperforms the compared algorithms especially in maximum the fitness value, so it can applied in image segmentation and optimization problems well.(4) Based on analysis of the convergence of particle swarm optimization (PSO), a new PSO based on improved Moderate Random Searching ability (IRPSO) is proposed to overcome the problem of bad searching ability in the last stage of the standard PSO. It helps the particles have more exploration ability and fast convergence rate. Furthermore, for the improved algorithm only having one parameter and iteration formula, it is simpler than PSO. Experiments show that the proposed algorithm performs much better than the other algorithms in terms of the quality of solution. For solving the problem in image segmentation, we use the difference of mutual information (DMI) as the optimization function, and the experiments show that the IRPSO algorithm gets the better performance of image segmentation among the compared algorithms.Finally, the work of this dissertation is summarized and the prospective of future research is discussed.