摘要
针对模糊C均值聚类算法中,聚类效果往往受到聚类中心数目和初始聚类中心的影响这一问题,提出一种基于多尺度自回归(MAR)模型与模糊C均值(FCM)聚类的声呐图像分割方法。引入MAR模型,建立层与层之间以及相邻层像素点间的数学关系,利用粗尺度图像的灰度-邻域均值二维直方图中的峰值个数来确定聚类中心数目,通过MAR得到的预测分割结果引导初始聚类中心的确定。实验结果表明,改进后的算法能准确、快速地确定聚类中心数目,并较好地解决初始聚类中心问题;与传统的FCM聚类方法相比,具有分割准确和收敛速度快的特点。
Aiming at the problem that the performance of fuzzy C-means (FCM) clustering algorithm depends on the selection of the number of cluster centers and initial cluster centers, this paper proposes a sonar image segmentation algorithm,which combines the multiscale auto regressive (MAR) model with the fuzzy FCM clustering algorithm. MAR model is used to establish the mathematic relationships among different layers of the image and the pixels of the adjacent layers. The number of cluster centers is determined by the peak number of the coarse scale image two-di- mensional histogram, which represents the relationship of the gray value of a pixel vs. the average gray value of the pixels within its neighborhood. The forecasted segmenting result of the initial image with the MAR model determines the initial cluster centers. Experimental results on a sonar image show that the proposed algorithm can accurately and quickly determine the cluster center number and appropriately solve the initial cluster center problem. Compared with traditional FCM clustering method, the algorithm features more accurate segmentation and faster convergence speed.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2013年第10期2322-2327,共6页
Chinese Journal of Scientific Instrument
基金
国家自然科学基金(41076060)
吉林省自然科学基金(20130101056JC)资助项目
关键词
多尺度自回归
模糊C均值聚类
声呐图像
二维直方图
muhiscale auto regressive (MAR)
fuzzy C-means clustering
sonar image
two-dimensional histogram