Abstract: Spectral clustering is used in pattern recognition extensively as a novel clustering algorithm in recent years.Due to the initialization dependence of original spectral clustering，this paper introduces the initialization insensitive k-harmonic means algorithm and proposes an initialization independent spectral clustering algorithm.Experiment on the artificial data set and the real data set shows that the improved algorithm has the remarkable enhancement in the stability and the clustering performance compared with traditional k-means algorithm，FCM algorithm and EM algorithm.

Key words: clustering, spectral clustering, k-harmonic means, initialization

摘要： 谱聚类作为一种新颖的聚类算法近年来受到模式识别领域的广泛关注。针对传统谱聚类算法对初始中心敏感的特点，通过引入对初值不敏感的k-调和平均算法，提出一种初始化独立的谱聚类算法。在人工数据和真实数据上的实验表明，相比于传统的k-means算法、FCM算法和EM算法，改进算法在稳定性和聚类性能上有了显著的提高。

关键词: 聚类, 谱聚类, k-调和平均, 初始化