Low-Rank Matrix Fitting Based on Subspace Perturbation Analysis with Applications to Structure from Motion

被引:23
作者
Jia, Hongjun [1 ]
Martinez, Aleix M. [1 ]
机构
[1] Ohio State Univ, Dept Elect & Comp Engn, Columbus, OH 43210 USA
基金
美国国家卫生研究院; 美国国家科学基金会;
关键词
Low-rank matrix; noise; missing data; random matrix; matrix perturbation; subspace analysis; structure from motion; computer vision; pattern recognition; MISSING DATA; PRINCIPAL COMPONENTS; SHAPE; RECOGNITION; ALGORITHM;
D O I
10.1109/TPAMI.2008.122
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
The task of finding a low-rank (r) matrix that best fits an original data matrix of higher rank is a recurring problem in science and engineering. The problem becomes especially difficult when the original data matrix has some missing entries and contains an unknown additive noise term in the remaining elements. The former problem can be solved by concatenating a set of r-column matrices that share a common single r-dimensional solution space. Unfortunately, the number of possible submatrices is generally very large and, hence, the results obtained with one set of r-column matrices will generally be different from that captured by a different set. Ideally, we would like to find that solution that is least affected by noise. This requires that we determine which of the r-column matrices (i.e., which of the original feature points) are less influenced by the unknown noise term. This paper presents a criterion to successfully carry out such a selection. Our key result is to formally prove that the more distinct the r vectors of the r-column matrices are, the less they are swayed by noise. This key result is then combined with the use of a noise model to derive an upper bound for the effect that noise and occlusions have on each of the r-column matrices. It is shown how this criterion can be effectively used to recover the noise-free matrix of rank r. Finally, we derive the affine and projective structure-from-motion (SFM) algorithms using the proposed criterion. Extensive validation on synthetic and real data sets shows the superiority of the proposed approach over the state of the art.
引用
收藏
页码:841 / 854
页数:14
相关论文
共 41 条
[1]   The computation of optical flow [J].
Beauchemin, SS ;
Barron, JL .
ACM COMPUTING SURVEYS, 1995, 27 (03) :433-467
[2]  
Brandt S., 2002, Proceedings of the Statistical Methods in Video Processing Workshop, P109
[3]  
Buchanan AM, 2005, PROC CVPR IEEE, P316
[4]   Recovering the missing components in a large noisy low-rank matrix: Application to SFM [J].
Chen, P ;
Suter, D .
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, 2004, 26 (08) :1051-1063
[5]  
De la Torre F, 2001, EIGHTH IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION, VOL I, PROCEEDINGS, P362, DOI 10.1109/ICCV.2001.937541
[6]  
Dodge Y., 1985, Analysis of experiments with missing data. Wiley Series in Probability and Mathematical Statistics
[7]   RANDOM SAMPLE CONSENSUS - A PARADIGM FOR MODEL-FITTING WITH APPLICATIONS TO IMAGE-ANALYSIS AND AUTOMATED CARTOGRAPHY [J].
FISCHLER, MA ;
BOLLES, RC .
COMMUNICATIONS OF THE ACM, 1981, 24 (06) :381-395
[8]  
Fitzgibbon A., 1998, LECT NOTES COMPUTER, V1506, P155
[9]  
Friedland S., 2006, PROC INT C ACOUSTICS, V2, P1092
[10]  
Golub G. H., 2012, Matrix computations, V4th