A Spectral Clustering Algorithm Based on Manifold Distance Kernel

For the problem that the similarity measure based on Euclidean distance cannot fully reflect the complex space distribution of data clustering in the standard spectral clustering algorithm, a novel spectral clustering algorithm is proposed based on manifold distance kernel. It can fully exploit the inherent structure information of the datasets. The proposed algorithm not only can reflect local and global consistency better, but also can prevent the influence of ''bridge" noise points, which improves the algorithm's clustering performance. Experimental results show that compared with traditional clustering algorithms and those popular spectral clustering algorithms, the algorithm can achieve better clustering effect on artificial datasets and UCI public datasets.