Divide-and-Conquer Method for L1 Norm Matrix Factorization in the Presence of Outliers and Missing Data

The low-rank matrix factorization as a L1 norm minimization problem has recently attracted much attention due to its intrinsic robustness to the presence of outliers and missing data. In this paper, we propose a new method, called the divide-and-conquer method, for solving this problem. The main idea is to break the original problem into a series of smallest possible sub-problems, each involving only unique scalar parameter. Each of these subproblems is proved to be convex and has closed-form solution. By recursively optimizing these small problems in an analytical way, efficient algorithm, entirely avoiding the time-consuming numerical optimization as an inner loop, for solving the original problem can naturally be constructed. The computational complexity of the proposed algorithm is approximately linear in both data size and dimensionality, making it possible to handle large-scale L1 norm matrix factorization problems. The algorithm is also theoretically proved to be convergent. Based on a series of experiment results, it is substantiated that our method always achieves better results than the current state-of-the-art methods on $L1$ matrix factorization calculation in both computational time and accuracy, especially on large-scale applications such as face recognition and structure from motion.

[1]  Donald E. Knuth,et al.  The Art of Computer Programming: Volume 3: Sorting and Searching , 1998 .

[2]  Yi Ma,et al.  The Augmented Lagrange Multiplier Method for Exact Recovery of Corrupted Low-Rank Matrices , 2010, Journal of structural biology.

[3]  Michael J. Black,et al.  A Framework for Robust Subspace Learning , 2003, International Journal of Computer Vision.

[4]  Nojun Kwak,et al.  Principal Component Analysis Based on L1-Norm Maximization , 2008, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[5]  Hongyuan Zha,et al.  {\it R}$_{\mbox{1}}$-PCA: rotational invariant {\it L}$_{\mbox{1}}$-norm principal component analysis for robust subspace factorization , 2006, ICML 2006.

[6]  Takeo Kanade,et al.  Robust L/sub 1/ norm factorization in the presence of outliers and missing data by alternative convex programming , 2005, 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'05).

[7]  David J. Kriegman,et al.  From Few to Many: Illumination Cone Models for Face Recognition under Variable Lighting and Pose , 2001, IEEE Trans. Pattern Anal. Mach. Intell..

[8]  John Wright,et al.  Robust Principal Component Analysis: Exact Recovery of Corrupted Low-Rank Matrices via Convex Optimization , 2009, NIPS.

[9]  Takayuki Okatani,et al.  On the Wiberg Algorithm for Matrix Factorization in the Presence of Missing Components , 2007, International Journal of Computer Vision.

[10]  Andrew W. Fitzgibbon,et al.  Damped Newton algorithms for matrix factorization with missing data , 2005, 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'05).

[11]  Michael J. Black,et al.  EigenTracking: Robust Matching and Tracking of Articulated Objects Using a View-Based Representation , 1996, International Journal of Computer Vision.

[12]  Gene H. Golub,et al.  Matrix computations , 1983 .

[13]  Tommi S. Jaakkola,et al.  Weighted Low-Rank Approximations , 2003, ICML.

[14]  Takeo Kanade,et al.  Shape and motion from image streams under orthography: a factorization method , 1992, International Journal of Computer Vision.

[15]  G. Sapiro,et al.  A collaborative framework for 3D alignment and classification of heterogeneous subvolumes in cryo-electron tomography. , 2013, Journal of structural biology.

[16]  Anders P. Eriksson,et al.  Efficient computation of robust low-rank matrix approximations in the presence of missing data using the L1 norm , 2010, 2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition.

[17]  Hideki Hayakawa Photometric stereo under a light source with arbitrary motion , 1994 .