中文摘要：

提出了一种基于多层次聚类的算法,此算法相对传统算法降低了空间和时间复杂度,并减少了对参数的敏感度,适合处理大规模数据集.该算法包括粗分、代表点聚类和细分三个阶段.首先,利用亲和传播聚类（AP）算法对所有数据进行粗分,为了节省空间和时间,仅考虑每个点和离自己最近的t个近邻之间的相似度,所以构造的相似度矩阵远远小于原始AP算法所构造的相似度矩阵;其次,为了进一步提高效率和性能,在第二阶段采用密度峰值算法（FDP）对上一阶段所得结果进行再划分;最后,结合两个阶段划分的结果得到所有数据的划分.实验表明：所提算法可以快速准确地进行图像分割,和经典聚类算法FCM（模糊C均值）、Kmeans以及SOM（自组织映射）的对比实验也证明了所提算法的有效性.

英文摘要：

A method based on multi-level clustering was proposed for saving the space and computa-tion and reducing the sensitivity of the parameters.This algorithm consists of three phases,which were coarsening,representative data clustering and exact partition.First,affinity propagation （AP） algorithm was used for coarsening.Specifically,in order to save the space and computational cost,on-ly the similarity between each point and its t nearest neighbors were computed,and a condensed simi-larity matrix was obtained.Second,to further improve the efficiency and effectiveness of the proposed algorithm,the find of density peaks （FDP）clustering was used to the resulted points （the representa-tive points gotten in the first phase）to do the representative data clustering.Third,the classes of all data were obtained by merging the results of the first two steps.As a result,the proposed algorithm can realize the clusters quickly and precisely for texture image segmentation.The experimental results show that the proposed algorithm is more efficient than the compared algorithms FCM （fuzzy C-means）,K-means and SOM （self-organizing maps）.