APP下载

基于动态粒子群优化与K-means聚类的图像分割算法

2018-05-15李立军张晓光

现代电子技术 2018年10期

李立军 张晓光

摘  要: 为了解决K?means聚类算法图像分割质量过度依赖于初始聚类中心选取,且易于陷入局部最优解等问题,提出一种基于动态粒子群优化(DPSO)与K?means聚类的图像分割算法(DPSOK)。通过动态调整惯性系数与学习因子来增强PSO算法的性能;然后计算粒子群适应度方差,找准切换至K?means算法时机;随后,将DPSO输出结果用来初始化K?means聚类中心,使其收敛至全局最优解;最后,通过最小化目标函数的多次迭代,使K?means的聚类中心不断更新,直到收敛。实验结果表明,DPSOK能有效提高K?means的全局搜索能力,在图像分割中它比K?means,PSO获得了更好的分割效果,且与粒子群优化和K?means算法相比, DPSOK算法具有更高的分割质量与效率。

关键词: 图像分割; 动态粒子群优化; K?means聚类; 适应度方差; 聚类算法; DPSOK

中图分类号: TN911.73?34; TP391          文献标识码: A                        文章编号: 1004?373X(2018)10?0164?05

Abstract: An image segmentation algorithm based on dynamic particle swarm optimization and K?means clustering (DPSOK) is proposed to resolve the problems that the image segmentation quality of K?means clustering algorithm overly relies on the selection of initial clustering center, and it is easy for the algorithm to fall into the local optimal solution. The performance of the particle swarm optimization (PSO) algorithm is enhanced by dynamically adjusting the inertia coefficient and the learning factor. The variance of the particle swarm adaptability is calculated, and the timing of switching to the K?means algorithm is captured. The output results of dynamic particle swarm optimization (DPSO) are used to initialize the K?means clustering center and enable it to converge to the global optimal solution. The K?means clustering center is updated constantly until reaching convergence by means of multiple iterations of the minimized objective function. The experimental results show that the DPSOK can effectively improve the global search capability of K?means, obtain a better segmentation effect than K?means and the PSO in image segmentation, and has higher segmentation quality and efficiency in comparison with the particle swarm optimization and K?means algorithm.

Keywords: image segmentation; dynamic particle swarm optimization; K?means clustering; fitness variance; clustering algorithm; DPSOK

0  引  言

K?means原理简单、计算速率高,已广泛应用于图像分割领域[1?3]。 但其也存在以下几个缺点[4]:K?means必须在算法初始化时给出聚类数k值;K?means对初始聚类中心选取要求很高;K?means容易收敛到局部最优解,将导致其错过全局最优解。

为了克服这些缺点,研究人员提出改进的K?means算法。如Chen提出基于层次的犹豫模糊K?means聚类算法[5],穆瑞辉提出了一种基于粒子群优化的模糊K?means目标分类算法 [6],Siddiqui提出一种增强型移动K?means聚类算法 [7]。但是,这些改进的算法的复杂度较高。因此,需要更进一步的研究来解决K?means的问题。另外,也有部分学者利用PSO算法[8?10]实现图像分割,但是PSO技术也存在着局部搜索能力较差、搜索精度不高并且容易陷入局部极值等缺点[9?10]。……

登录APP查看全文