Spectral Clustering for Robust Motion Segmentation

In this paper, we propose a robust motion segmentation method based on the matrix factorization and subspace separation. We, first, mathematically prove that the shape interaction matrix can be derived using QR decomposition rather than Singular Value Decomposition(SVD). Using shape interaction matrix, we solve the motion segmentation problem using spectral graph clustering technique. We exploit K-way Min-Max cut clustering method. This method yields a good performance for noise-free data, but it is sensitive to noise. We combine a cluster refinement method based on subspace separation to the spectral graph clustering to deal with noise. The proposed method yields very good performance for both synthetic and real image sequences.