中文摘要：

针对传统的K均值聚类算法随机选取初始聚类中心与分类类别数的缺陷,提出了一种新的初始聚类中心与分类类别数的选取方法,并将此方法应用在彩色图像有意义区域提取中。实验证实：新算法不仅能有效地改善初始聚类中心,而且能够提高图像分割的精度;与复杂的协方差矩阵分割算法相比,算法更易于实现有意义区与背景的分离,分割效果令人满意。

英文摘要：

Considering the shortcoming of traditional K-means clustering algorithm in random selection of initial clustering centre and category number of classification, a new approach for selecting initial clustering centre and category number of classification is proposed in this pa- per, and this new algorithm has been applied to extracting meaningful region of colour image. Experimental results prove that the new algorithm can meliorate initial clustering centre effectively as well as improve image segmentation precision. Compared with the complicated covariance matrix segmentation algorithm, the new improved K-means algorithm can more easily realize the separation of the background and the meaningful region, and the segmentation results are satisfied.