Research Of Spectral Clustering Based On Density And Path

In recent years, spectral clustering algorithm is a significant branch of clustering analysis, is one of the main research topics in subject areas like pattern recognition, machine learning, data mining and so on. Spectral clustering is one kind of clustering algorithms, which is based on similarity matrix, and divides the similarity matrix according to spectral analysis theory. Its essence is point to point clustering method that converts clustering problem into graph optimal partition problem. Unlike the traditional clustering algorithms, spectral clustering can solve the clustering of non-convex sphere of sample spaces, So that it can be converged to global optimal solutionThis paper does one study on density-path-based spectral clustering algorithm, which is based on the study of traditional spectral clustering algorithm, density-based clustering algorithm and path-based clustering algorithm then combines the advantages of density-based clustering algorithm and path-based clustering algorithm. The studied algorithm is mainly to construct a new similarity matrix, and the new similarity matrix is approximate to the ideal block diagonal similarity matrix according to experiments. This paper gives some experiments on the synthetic datasets and handwriting databases, the clustering results reveal that this algorithm has more than a little development in clustering performance than the traditional spectral clustering.This paper also studies the robust density-path-based spectral clustering algorithm, which is based on the study of density-path-based spectral clustering and adds robust coefficient by neighborhood information of the database to enhance its anti-disturbance performance. This paper also compares the studied algorithm with the DBSCAN algorithm, the PB-SC algorithm and the RPB-SC algorithm in clustering performance through some experiments. The experiment results on synthetic datasets and handwriting databases show that the robust density-path-based spectral clustering algorithm outperforms the other three algorithms in clustering performance.