关键词: 聚类, 谱聚类, k-means, 多尺度, 特征向量

Abstract: A multiscale spectral clustering algorithm is proposed.Unlike the traditional spectral clustering algorithm，multiscale spectral clustering algorithm uses a modified k-means algorithm to cluster unstandardized Laplacian matrix eigenvector.Unlike the traditional k-means algorithm，the improved algorithm proposes a novel method to partition data points to cluster centers，which calculates the distance of data points and cluster centers through comparing the distance of cluster centers and origin and introducing scale parameters.Experiments show that it can acquire satisfactory results on artificial data sets，but also it can get better cluster results on real data sets.

Key words: clustering, spectral clustering, k-means, multiscale, eigenvector