Intelligent Algorithm For Image Segmentation Groups Performance Analysis And Optimization Studies

Swarm intelligence is a new approach used to solve some combined optimization problems, which has been paid much attention to by the researchers and brought plenty of application fruits due to the outstanding ability to find solutions in parallel. However, most of optimization methods based on swarm intelligence are still lacking of theoretical guidance in image processing, in which many controlling parameters are selected on experience. Additionally, different swarms are always moving based on different individual modes and various random, which result in the difference of performance when dealing with digital images.This dissertation titled as "Research of Performance Analysis and Optimization Methods of Swarm Intelligence based on Image Segmentation", will focus on two aspects. On one hand, based on the theoretical analysis of exsting optimization algorithms, effective performance evaluation of optimization algorithms based on swarm intelligence are researched and the performance of the algorithms based on image segmentation are compared. On the other hand, some basic optimization algorithms are impoved or combined with some algorithms to improve the algorithm performance when used in image segmentation. The main contributions are summarizes as follows:(1) A comparative platform is established for optimization algorithms based on swarm intelligence, some principles of parameter selection are determined to compare their performance, and some new evaluation indexes in image segmentation are also suggested based on stable convergence generation, grey relational degree of convergence and convergence area. Then, considering two-dimensional Otsu as the fitness function, the performance of genetic algorithm, particle swarm optimization algorithm, artificial fish swarm algorithm, bacterial foraging algorithm and artificial bee colony algorithm are objectively evaluated. Experimental results show that algorithms adopted the information of the best or better individuals'have better performance.(2) After a deep analysis of basic behavior characteristics of bacterial foraging algorithm, an improved bacterial foraging algorithm is proposed to solve the problem of difficultly finding the optimal solution in large step. In the improved bacterial foraging algorithm, each bacterium can make use of the optimal individual's information in chemotaxis behavior in large step so that the algorithm can find the optimal solution quickly and accurately. Experiments in image segmentation show that the improved bacterial foraging algorithm is better than particle swarm optimization algorithm, artificial bee colony algorithm and basic bacterial foraging algorithm in terms of the stable convergence generation, convergence area and relational degree of convergence, namely, the improved bacterial foraging algorithm has the best convergence performance and is the most stable.(3) After a deep analysis of basic behavior characteristics of artificial fish swarm algorithm, the basic artificial fish swarm algorithm is combined with the ability of continuous forward in chemotaxis behavior of bacterial foraging algorithm. In the improved artificial fish swarm algorithm, each fish has the ability of continuous forward in its3basic kinds of behavior patterns, which further promotes the ability of fast moving to the better direction. Experiments in image segmentation show that the improved artificial fish swarm algorithm is better than bacterial foraging algorithm, artificial bee colony algorithm, particle swarm optimization algorithm, and basic artificial fish swarm algorithm in terms of the stable convergence generation, convergence area and relational degree of convergence, namely, the improved artificial fish swarm algorithm has the best convergence performance and is the most stable.(4) After a deep analysis of basic behavior characteristics of artificial bee colony, the basic artificial bee colony is modified on the pattern of finding new nectar posions of the employed bees and onlookers, which reduces the probability of finding worse nectar posions. Experiments in image segmentation show that the improved artificial bee colony algorithm is better than bacterial foraging algorithm and basic artificial bee colony algorithm in terms of the stable convergence generation, convergence area and relational degree of convergence, namely, the improved artificial fish swarm algorithm has the best convergence performance is the most stable.