基于改进K-均值聚类的图像分割算法研究

Research of K-means Clustering Algorithm for Images Based on Weighted Euclidean Distance

为了实现彩色图像的准确分割,研究了在HLS颜色空间中基于优化初始中心的加权K-均值彩色图像聚类算法.首先对大样本的目标颜色进行数理统计,获取优化的初始聚类中心,从而实现准确分类和避免K-均值容易陷入局部最优的问题;然后在HLS颜色空间中引入加权欧氏距离来度量对象间的相关性,通过调整系数使对象不同的颜色属性内在特征得以充分利用.实验证明,该算法在保持K-均值聚类简洁、收敛速度快的同时能产生更好的聚类效果,实现彩色图像的快速准确分割.

In order to achieve accurate segmentation for color image,a K-means clustering algorithm for color image based on weighted Euclidean distance was studied in HLS color space.It obtained the optimized initial clustering centers which were computed from the mathematical statistics results of large pictures to resolve the problems that the clustering results of K-means were likely to be trapped into local optimums firstly.Then the weighted Euclidean distance by use of the different color properties was introduced to measure the relevance between objects in HLS color space.Experimental results showed this algorithm could take advantage of the concise and fast convergence features of K-means clustering,and produce more better clustering results to realized the accurate color image segmentation.