摘要
提出一种改进的基于遥感图像的颜色和纹理特征进行聚类的K均值算法。该算法通过统计图像色度直方图的峰值,来获得三组聚类个数和初始聚类中心,并结合色度和基于灰度共生矩阵的纹理特征形成图像聚类特征,然后进行改进的K均值聚类,最后选择silhouette均值最大的一组作为最佳聚类结果。该方法的随机性和聚类误差比传统K均值算法小,实验结果证实了该方法的可行性和有效性。
This paper presents an improved K-means Clustering algorithm based on colour and texture feature of the remote sensing images. By the statistics of peak value of the hue histogram, the algorithm gets three groups of the number of clusters and initial centre points. Clustering feature of image is composed in combination of hue and texture feature which is based on grey-level co-occurrence matrix. Then the improved K-means clustering is performed, and the group ,which has the maximum of silhouette mean, is chosen as the best clustering result. The algorithm has less randomicity and clustering error than traditional K-means method. The results of experiments show the feasibility and effectiveness of it.
出处
《计算机应用与软件》
CSCD
2009年第11期246-248,285,共4页
Computer Applications and Software
