An Optimal Spectral Clustering Approach Based on Cauchy-Schwarz Divergence

A new global criterion, the CauchySchwarz (CS) cut, for spectral clustering is presented basedon the Cauchy-Schwarz divergence. It is proved thatwhen the sum of intra-cluster and inter-cluster similarity is fixed, optimizing the CS cut criterion can ensureintra-cluster similarity maximized and inter-cluster similarity minimized simultaneously. An effcient computational technique is developed based on eigenvalue problem.Experimental results on artificial data sets and natural images show that the proposed approach is very encouraging.