A Spectral Clustering Algorithm Based on Normalized Cuts

Recently, spectral clustering has wide application in pattern recognition and data mining because it can obtain global optima solution and adapt to sample spaces with any shape. Thus, a spectral clustering algorithm based on normalized cuts is proposed in this paper. It selects the k eigenvalues and corresponding eigenvectors of a given stochastic matrix and clusters in n times k sub-space. Experimental results show that it has better performance comparing with the traditional clustering algorithm.