Improved Particle Swarm Optimization Algorithm And Its Application In Image Segmentation

Particle swarm optimization (PSO) algorithm is a novel evolutional optimizing method which is inspired by the social behaviors of bird flocks and fish schools. It has many advantages, such as simple principles, few parameters, easy implement, and high efficiency. PSO algorithm has become a hot spot which attracts more and more researchers all around the world to study and explore. Recently PSO algorithm and the improved models have been widely applied to solve the optimizing problems of many fields, like multi-objective optimization, pattern recognition, data mining and image segmentation. However, owing to the imperfectness of PSO algorithm in theory analysis and practical application, there are a large number of problems left to make further study.This paper profoundly analyzes the theories and applications of PSO algorithm, the main work are as follows:Firstly, aiming to the drawbacks of premature pheromone and slow convergence in the last phase of evolution of PSO algorithm, this paper proposes two improved methods:one is dividing the updated formula of speed into three sections on the basis of Geese-PSO algorithm. The former two sections are considered as the personal information and the last section is regarded as the global information. The improved algorithm mainly focuses on respectively adding weight factors on the personal information and global information which guarantees that the particles barely take personal information into account in the former stage and only attach importance on sharing the global information with each other in the latter stage. The advantages of doing this are to improve the global searching ability in the former stage and make the local searching ability high to avoid premature convergence in the latter. In order to escape from the excessive homoplasy in the last stage, the algorithm also combines the random disturbed strategy to help the particles jump out of local optimum values. Another improvement is to blend the artificial bee colony (ABC) with PSO algorithm by means of the neighbor searching ideas of scouts in ABC algorithm. The principle is that initially each individual is assigned with a neighbor space for itself according to some rules. Then search the optimum solution in the proximity space, if the personal best in the neighbor space is better than the current global best, then the algorithm substitutes the personal best in the neighbor space instead. The advantage of searching in the neighbor space of particles is to maintain the varieties of swarms. Secondly, this paper also discusses the combination between PSO algorithm and the spectral clustering algorithm based on graph-mapping theory. First, PSO algorithm has the features of strong searching ability and quick convergent speed. Second, spectral clustering algorithm is able to deal with high throughput data with various shapes and reduce dimensions at the same time. Therefore this paper tentatively adopts PSO algorithm to optimize the k-means clustering method after obtaining the former k eigenvectors of Laplacian matrix in the spectral clustering algorithm. The ultimate goal is to improve the executing efficiency of k-means clustering on one hand. On the other hand, the improved algorithm can efficiently solve several high-dimensional problems such as functional optimization, engineering optimization and so on.In the end, this paper introduces the improved PSO algorithms into image segmentation. The two dimension maximum entropy threshold segmentation is one of the relatively mature segmentation methods in the theory system of image segmentation. The key of this method is to determine a reasonable and proper threshold quickly and efficiently. This paper takes advantage of the quick convergence and high searching accuracy of improved PSO algorithm to search the optimum threshold. The simulated results turn out that the algorithm not only efficiently and accurately searches optimum segmentation threshold, but also obtains a relatively satisfying segmentation effect.