Proximal Plane Clustering via Eigenvalues

By replacing the clustering center by plane, k-plane clustering has been studied wildly. In this paper, we propose a proximal plane clustering, which the clustering plane in k-plane clustering is improved to be such a plane that is closer to the objective data points and far away from the others, which leads to solve the eigenvalue problems. Therefore, our proximal plane clustering can handle many cases, especially the data points are almost on the same hyperplane, while k-plane clustering fails to these problems. Experimental results on several visual and public available datasets show the better performance of the proximal plane clustering than k-plane clustering.