Point cloud segmentation based on spectral clustering

A spectral clustering based method is proposed to segment point cloud into meaningful subparts. By representing the point cloud as a graph G, the segmentation problem can be turned into a graph min-cut problem. The nonsymmetric normalized Laplacian matrix is used to construct the spectral space. By removing redundant eigenvectors from the spectral domain, the segmentation solution is found in a lower dimensional space. The theoretical guarantee of the proposed method is proved. The accuracy and efficiency of the algorithm are verified by experimental results.