Spectral Clustering Method and Its Application

Abstract

Despite many empirical successes of spectral clustering methods which use eigenvectors of matrices derived from the data, this paper studies automatic segmentation of multiple motions or patterns from the tracked feature points. We propose an affinity matrix definition algorithm and suggest upper bounds together with a data-driven procedure for choosing automatically the optimal cluster number for the spectral clustering. Our approach is applied to the video benchmark databases and shows good experimental results on a number of clustering problems, such as motion pattern segmentation.