Spectral clustering algorithm based on K-nearest neighbor measure

Analysing the defect on different similarity matrix in spectral clustering, we propose a new algorithm—Spectral clustering algorithm based on K-nearest neighbor measure. The K-nearest neighbor measure focuses on using data points between the common number of nearest neighbors to measure the degree of similarity, and avoids the degree of similarity is large and unstable by contrast. The experiment results show the efficiency and performance of the algorithm. Meanwhile, the algorithm effectively resolves the problem that two data points belonging to different clusters are close. It possesses the advantage of discriminating the clusters with variable density, and also has the advantage of clustering in a sample space of any shape.