Bilinear Factor Matrix Norm Minimization for Robust PCA: Algorithms and Applications
Open accessThe DOI matches its Crossref record; 74 of the 74 references it cites were found in a registry or an index.
Abstract
The heavy-tailed distributions of corrupted outliers and singular values of all channels in low-level vision have proven effective priors for many applications such as background modeling, photometric stereo and image alignment. And they can be well modeled by a hyper-Laplacian. However, the use of such distributions generally leads to challenging non-convex, non-smooth and non-Lipschitz problems, and makes existing algorithms very slow for large-scale applications. Together with the analytic solutions to $\ell _{p}$ -norm minimization with two specific values of $p$ , i.e., $p=1/2$ and $p=2/3$ , we propose two novel bilinear factor matrix norm minimization models for robust principal component analysis. We first define the double nuclear norm and Frobenius/nuclear hybrid norm penalties, and then prove that they are in essence the Schatten- $1/2$ and $2/3$ quasi-norms, respectively, which lead to much more tractable and scalable Lipschitz optimization problems. Our experimental analysis shows that both our methods yield more accurate solutions than original Schatten quasi-norm minimization, even when the number of observations is very limited. Finally, we apply our penalties to various low-level vision problems, e.g., text removal, moving object detection, image alignment and inpainting, and show that our methods usually outperform the state-of-the-art methods.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Golub, G. (2012) · doi:10.56021/9781421407944 · Cited by 34,505
A theory for multiresolution signal decomposition: the wavelet representation
Mallat, S. G. (1989) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/34.192463 · Cited by 21,372 · cited 1×
“8, which correspond to non-convex lpnorms) [10], [11], as exhibited by the sparse noise/outliers in low-level vision problems [12] shown in Fig.”
BackgroundBoyd, S., Parikh, N., Chu, E., Peleato, B., & Eckstein, J. (2011) · Foundations and Trends® in Machine Learning · doi:10.1561/2200000016 · Cited by 17,075
Boyd, S. (2010) · now publishers, Inc. eBooks · doi:10.1561/9781601984616 · Cited by 12,677
A tutorial on spectral clustering
von Luxburg, U. (2007) · Statistics and Computing · doi:10.1007/s11222-007-9033-z · Cited by 10,349
De-noising by soft-thresholding
Donoho, D. L. (1995) · IEEE Transactions on Information Theory · doi:10.1109/18.382009 · Cited by 9,619 · cited 1×
“This is mainly because the optimal solutions of the sub-problems involving both terms in (3) can be obtained by two wellknown proximal operators: the singular value thresholding (SVT) operator [23] and the soft-thresholding operator [52].”
Background$rm K$-SVD: An Algorithm for Designing Overcomplete Dictionaries for Sparse Representation
Aharon, M., Elad, M., & Bruckstein, A. (2006) · IEEE Transactions on Signal Processing · doi:10.1109/tsp.2006.881199 · Cited by 9,451 · cited 1×
“THE sparse and low-rank priors have been widely used in many real-world applications in computer vision and pattern recognition, such as image restoration [1], face recognition [2], subspace clustering [3], [4], [5] and robust principal component analysis [6] (RPCA, also called low-rank and sparse matrix decomposition…”
BackgroundRobust Face Recognition via Sparse Representation
Wright, J., Yang, A. Y., Ganesh, A., Sastry, S. S., & Ma, Y. (2009) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2008.79 · Cited by 9,388 · cited 1×
“THE sparse and low-rank priors have been widely used in many real-world applications in computer vision and pattern recognition, such as image restoration [1], face recognition [2], subspace clustering [3], [4], [5] and robust principal component analysis [6] (RPCA, also called lowrank and sparse matrix decomposition…”
BackgroundVariable Selection via Nonconcave Penalized Likelihood and its Oracle Properties
Fan, J., & Li, R. (2001) · Journal of the American Statistical Association · doi:10.1198/016214501753382273 · Cited by 8,900 · cited 1×
“However, it has been shown in [18] that the l1-norm over-penalizes large entries of vectors and results in a biased solution.”
BackgroundRobust principal component analysis?
Candès, E. J., Li, X., Ma, Y., & Wright, J. (2011) · Journal of the ACM · doi:10.1145/1970392.1970395 · Cited by 6,921 · cited 9×
“In this section, we mainly discuss some recent advances in RPCA, and briefly review some existing work on RPCA and its applications in computer vision (readers may see [6] for a review).”
BackgroundMethodInfluentialA Singular Value Thresholding Algorithm for Matrix Completion
Cai, J.-F., Candès, E. J., & Shen, Z. (2010) · SIAM Journal on Optimization · doi:10.1137/080738970 · Cited by 5,809 · cited 4×
“To solve such a problem, the rank function is usually relaxed by its convex envelope [5], [23], [24], [25], [26], [27], [28], [29], the nuclear norm (i.”
BackgroundMethodInfluentialAn iterative thresholding algorithm for linear inverse problems with a sparsity constraint.
Daubechies, I., Defrise, M., & De Mol, C. (2004) · Communications on Pure and Applied Mathematics · doi:10.1002/cpa.20042 · Cited by 4,958
Nearly unbiased variable selection under minimax concave penalty
Zhang, C.-H. (2010) · The Annals of Statistics · doi:10.1214/09-aos729 · Cited by 3,835 · cited 1×
“Right: the empirical distribution of the sparse component (blue solid line), along with a hyper-Laplacian fit with α =1 / 2 (green dashdot line) and α =2 / 3 (red dotted line). many non-convex surrogates of the (cid:96) 0 -norm listed in [19] give a closer approximation, e.g., SCAD [18] and MCP [20].”
BackgroundRobust Recovery of Subspace Structures by Low-Rank Representation
Liu, G., Lin, Z., Yan, S., Sun, J., Yu, Y., & Ma, Y. (2013) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2012.88 · Cited by 3,594 · cited 4×
“To solve such a problem, the rank function is usually relaxed by its convex envelope [5], [23], [24], [25], [26], [27], [28], [29], the nuclear norm (i.”
BackgroundMethodInfluentialSelf-citationGuaranteed Minimum-Rank Solutions of Linear Matrix Equations via Nuclear Norm Minimization
Recht, B., Fazel, M., & Parrilo, P. A. (2010) · SIAM Review · doi:10.1137/070697835 · Cited by 3,471 · cited 5×
“[15] replaced the nuclear norm regularizer in (3) with the equivalent non-convex formulation stated in Lemma 1, and proposed scalable bilinear spectral regularized models, similar to collaborative filtering applications in [26], [67].”
BackgroundMethodInfluentialSparse Subspace Clustering: Algorithm, Theory, and Applications
Elhamifar, E., & Vidal, R. (2013) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2013.57 · Cited by 3,106
Weighted Nuclear Norm Minimization with Application to Image Denoising
Gu, S., Zhang, L., Zuo, W., & Feng, X. (2014) · 2014 IEEE Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2014.366 · Cited by 2,316 · cited 3×
“It has been shown in [36] that the singular values of nonlocal matrices in natural images usually exhibit a heavytailed distribution (p(σ)∝ e−k|σ|α ), as well as the similar phenomena in natural scenes [37], [38], as shown in Fig.”
BackgroundMethodKuybeda, O., Frank, G. A., Bartesaghi, A., Borgnia, M., Subramaniam, S., & Sapiro, G. (2013) · Journal of Structural Biology · doi:10.1016/j.jsb.2012.10.010 · Cited by 2,162 · cited 9×
“We also compare our methods with several state-of-the-art methods, such as LMaFit3 [47], RegL14 [16], Unifying [49], factEN5 [15], RPCA6 [44], PSVT7 [45], WNNM8 [46], and LpSq9 [31].”
BackgroundMethodInfluentialProximal alternating linearized minimization for nonconvex and nonsmooth problems
Bolte, J., Sabach, S., & Teboulle, M. (2014) · Mathematical Programming · doi:10.1007/s10107-013-0701-9 · Cited by 1,688 · cited 2×
“, ∂f(x) = {u ∈ R : ∃x → x, f(x)→ f(x) and u ∈ ∂̂f(x)→ u as k→∞}, and ∂̂f(x) is the Frèchet sub-differential of f at x (see [75] for more details).”
Regularization methods for learning incomplete matrices
Mazumder, R., Hastie, T. J., & Tibshirani, R. J. (2009) · Journal of Machine Learning Research · doi:10.48550/arxiv.0906.2034 · Cited by 1,645
Wright, J. N., Ganesh, A., Rao, S., Peng, Y., & Ma, Y. (2009) · Illinois Digital Environment for Access to Learning and Scholarship (University of Illinois at Urbana-Champaign) · Cited by 1,495
Fast Image Deconvolution using Hyper-Laplacian Priors
Krishnan, D., & Fergus, R. (2009) · neural information processing systems · Cited by 1,470
The Augmented Lagrange Multiplier Method for Exact Recovery of Corrupted Low-Rank Matrices
Lin, Z., Chen, M., & Ma, Y. (2010) · Illinois Digital Environment for Access to Learning and Scholarship (University of Illinois at Urbana-Champaign) · doi:10.48550/arxiv.1009.5055 · Cited by 1,176 · cited 9×
“We also compare our methods with several state-of-the-art methods, such as LMaFit3 [47], RegL14 [16], Unifying [49], factEN5 [15], RPCA6 [44], PSVT7 [45], WNNM8 [46], and LpSq9 [31].”
BackgroundMethodInfluentialSelf-citation$L_{1/2}$ Regularization: A Thresholding Representation Theory and a Fast Solver
Xu, Z., Chang, X., Xu, F., & Zhang, H. (2012) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2012.2197412 · Cited by 1,062 · cited 3×
“Before giving the proof of Proposition 1, we first give the following lemma [21].”
BackgroundMethodMaximum-Margin Matrix Factorization
Srebro, N., Rennie, J. D. M., & Jaakkola, T. (2004) · Cited by 962
Toh, K.-C., & Yun, S. (2009) · National University of Singapore · Cited by 955
Hong, M., Luo, Z.-Q., & Razaviyayn, M. (2016) · SIAM Journal on Optimization · doi:10.1137/140990309 · Cited by 841 · cited 1×
“As stated in [45], [76], the general convergence property of the ADMM for non-convex problems has not been answered yet, especially for multi-block cases [76], [77].”
BackgroundSelf-citationFast and Accurate Matrix Completion via Truncated Nuclear Norm Regularization
Hu, Y., Zhang, D., Ye, J., Li, X., & He, X. (2013) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2012.271 · Cited by 829 · cited 8×
“It has been shown in [36] that the singular values of nonlocal matrices in natural images usually exhibit a heavytailed distribution (p(σ)∝ e−k|σ|α ), as well as the similar phenomena in natural scenes [37], [38], as shown in Fig.”
BackgroundMethodResultInfluentialRASL: Robust Alignment by Sparse and Low-Rank Decomposition for Linearly Correlated Images
Peng, Y., Ganesh, A., Wright, J., Xu, W., & Ma, Y. (2012) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2011.282 · Cited by 817
Weighted Nuclear Norm Minimization and Its Applications to Low Level Vision
Gu, S., Xie, Q., Meng, D., Zuo, W., Feng, X., & Zhang, L. (2017) · International Journal of Computer Vision · doi:10.1007/s11263-016-0930-5 · Cited by 816 · cited 6×
“We also compare our methods with several state-of-the-art methods, such as LMaFit3 [47], RegL14 [16], Unifying [49], factEN5 [15], RPCA6 [44], PSVT7 [45], WNNM8 [46], and LpSq9 [31].”
BackgroundMethodInfluentialChen, C., He, B., Ye, Y., & Yuan, X. (2014) · Mathematical Programming · doi:10.1007/s10107-014-0826-5 · Cited by 793 · cited 1×
“As stated in [45], [76], the general convergence property of the ADMM for non-convex problems has not been answered yet, especially for multi-block cases [76], [77].”
BackgroundRank-Sparsity Incoherence for Matrix Decomposition
Chandrasekaran, V., Sanghavi, S., Parrilo, P. A., & Willsky, A. S. (2011) · SIAM Journal on Optimization · doi:10.1137/090761793 · Cited by 686 · cited 2×
“To solve such a problem, the rank function is usually relaxed by its convex envelope [5], [23], [24], [25], [26], [27], [28], [29], the nuclear norm (i.”
BackgroundSpectral Regularization Algorithms for Learning Large Incomplete Matrices
MazumderRahul, HastieTrevor, & TibshiraniRobert (2010) · Journal of Machine Learning Research · doi:10.5555/1756006.1859931 · Cited by 671 · cited 1×
“Because the trace of the product of two PSD matrices is always non-negative (see the proof of Lemma 6 in [73]), we have Tr In LUΣ 12 U LU”
BackgroundMoving Object Detection by Detecting Contiguous Outliers in the Low-Rank Representation
Zhou, X., Yang, C., & Yu, W. (2013) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2012.132 · Cited by 648 · cited 5×
“To induce sparsity, a principled way is to use the convex l1-norm [6], [13], [14], [15], [16], [17], which is the closest convex relaxation of the sparser lp-norm, with compressed sensing being a prominent example.”
BackgroundMethodInfluentialA Framework for Robust Subspace Learning
De la Torre, F., & Black, M. J. (2003) · International Journal of Computer Vision · doi:10.1023/a:1023709501986 · Cited by 604 · cited 1×
“THE sparse and low-rank priors have been widely used in many real-world applications in computer vision and pattern recognition, such as image restoration [1], face recognition [2], subspace clustering [3], [4], [5] and robust principal component analysis [6] (RPCA, also called lowrank and sparse matrix decomposition…”
BackgroundKe, Q., & Kanade, T. (2005) · 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'05) · doi:10.1109/cvpr.2005.309 · Cited by 578 · cited 1×
“To induce sparsity, a principled way is to use the convex (cid:96) 1 -norm [6], [13], [14], [15], [16], [17], which is the closest convex relaxation of the sparser (cid:96) p -norm, with compressed sensing being a prominent example.”
MethodRobust video denoising using low rank matrix completion
Ji, H., Liu, C., Shen, Z., & Xu, Y. (2010) · 2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2010.5539849 · Cited by 483 · cited 1×
“The formulation (3) has been widely used in many computer vision applications, such as object detection and background subtraction [17], image alignment [50], low-rank texture analysis [29], image and video restoration [51], and subspace clustering [27].”
MethodRecovering Low-Rank and Sparse Components of Matrices from Incomplete and Noisy Observations
Tao, M., & Yuan, X. (2011) · SIAM Journal on Optimization · doi:10.1137/100781894 · Cited by 453 · cited 1×
“THE sparse and low-rank priors have been widely used in many real-world applications in computer vision and pattern recognition, such as image restoration [1], face recognition [2], subspace clustering [3], [4], [5] and robust principal component analysis [6] (RPCA, also called lowrank and sparse matrix decomposition…”
BackgroundLaplacian Regularized Low-Rank Representation and Its Applications
Yin, M., Gao, J., & Lin, Z. (2016) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2015.2462360 · Cited by 442
Self-citationA Multimedia Retrieval Framework Based on Semi-Supervised Ranking and Relevance Feedback
Yang, Y., Nie, F., Xu, D., Luo, J., Zhuang, Y., & Pan, Y. (2011) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2011.170 · Cited by 417 · cited 1×
“In addition, we are interested in exploring ways to regularize our models with auxiliary information, such as graph Laplacian [27], [80], [81] and hyper-Laplacian matrix [82], or the elastic-net [15].”
MethodA Generalized Iterated Shrinkage Algorithm for Non-convex Sparse Coding
Zuo, W., Meng, D., Zhang, L., Feng, X., & Zhang, D. (2013) · 2013 IEEE International Conference on Computer Vision · doi:10.1109/iccv.2013.34 · Cited by 396 · cited 4×
“Before giving the proof of Proposition 2, we first give the following lemma [22], [74].”
BackgroundMethodInfluentialImage Clustering Using Local Discriminant Models and Global Integration
Yang, Y., Xu, D., Nie, F., Yan, S., & Zhuang, Y. (2010) · IEEE Transactions on Image Processing · doi:10.1109/tip.2010.2049235 · Cited by 389 · cited 1×
“In addition, we are interested in exploring ways to regularize our models with auxiliary information, such as graph Laplacian [27], [80], [81] and hyper-Laplacian matrix [82], or the elastic-net [15].”
MethodImproved Iteratively Reweighted Least Squares for Unconstrained Smoothed $\ell_q$ Minimization
Lai, M.-J., Xu, Y., & Yin, W. (2013) · SIAM Journal on Numerical Analysis · doi:10.1137/110840364 · Cited by 385
A trace inequality of John von Neumann
Mirsky, L. (1975) · Monatshefte für Mathematik · doi:10.1007/bf01647331 · Cited by 343 · cited 7×
“Let X ∈ R n × n be a symmetric positive semi-definite (PSD) matrix and its full SVD be Lemma 2 can be seen as a special case of the well-known von Neumann’s trace inequality, and its proof is provided in the Supplementary Materials.”
MethodInfluentialOnline Robust PCA via Stochastic Optimization
Feng, J., Xu, H., & Yan, S. (2013) · National University of Singapore · Cited by 310
Guaranteed Matrix Completion via Non-Convex Factorization
Sun, R., & Luo, Z.-Q. (2016) · IEEE Transactions on Information Theory · doi:10.1109/tit.2016.2598574 · Cited by 309 · cited 1×
“The bilinear spectral penalty in the first equality of (4) has been widely used in low-rank matrix completion and recovery problems, such as RPCA [15], [49], online RPCA [68], matrix completion [69], and image inpainting [25].”
MethodSelf-citationShen, Y., Wen, Z., & Zhang, Y. (2014) · Optimization Methods and Software · doi:10.1080/10556788.2012.700713 · Cited by 299 · cited 11×
“Given that d m;n, the overall complexity of Algorithm 1, as well as Algorithm 2, is thus OðmndÞ, which is the same as the complexity of LMaFit [47], RegL1 [16], ROSL [64], Unifying [49], and factEN [15].”
BackgroundMethodInfluentialGeneralized Transfer Subspace Learning Through Low-Rank Constraint
Shao, M., Kit, D., & Fu, Y. (2014) · International Journal of Computer Vision · doi:10.1007/s11263-014-0696-6 · Cited by 298 · cited 1×
“The l1-norm penalty in (3) can also be replaced by the l1,2norm as in outlier pursuit [28], [53], [54], [55] and subspace learning [5], [56], [57].”
BackgroundTILT: Transform Invariant Low-Rank Textures
Zhang, Z., Ganesh, A., Liang, X., & Ma, Y. (2012) · International Journal of Computer Vision · doi:10.1007/s11263-012-0515-x · Cited by 277 · cited 2×
“To solve such a problem, the rank function is usually relaxed by its convex envelope [5], [23], [24], [25], [26], [27], [28], [29], the nuclear norm (i.”
BackgroundMethodGeneralized Nonconvex Nonsmooth Low-Rank Minimization
Lu, C., Tang, J., Yan, S., & Lin, Z. (2014) · 2014 IEEE Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2014.526 · Cited by 271 · cited 5×
“Moreover, both our methods are much faster than the other methods, in particular, more than 25 times faster than the methods [19], [38], [46].”
BackgroundMethodInfluentialSelf-citationPartial Sum Minimization of Singular Values in Robust PCA: Algorithm and Applications
Oh, T.-H., Tai, Y.-W., Bazin, J.-C., Kim, H., & Kweon, I. S. (2016) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2015.2465956 · Cited by 238 · cited 10×
“We also compare our methods with several state-of-the-art methods, such as LMaFit3 [47], RegL14 [16], Unifying [49], factEN5 [15], RPCA6 [44], PSVT7 [45], WNNM8 [46], and LpSq9 [31].”
BackgroundMethodResultInfluentialUnifying Nuclear Norm and Bilinear Factorization Approaches for Low-Rank Matrix Decomposition
Cabral, R., De la Torre, F., Costeira, J. P., & Bernardino, A. (2013) · 2013 IEEE International Conference on Computer Vision · doi:10.1109/iccv.2013.309 · Cited by 204 · cited 10×
“The bilinear spectral penalty in the first equality of (4) has been widely used in low-rank matrix completion and recovery problems, such as RPCA [15], [49], online RPCA [68], matrix completion [69], and image inpainting [25].”
BackgroundMethodInfluentialFeature Selection for Multimedia Analysis by Sharing Information Among Multiple Tasks
Yang, Y., Ma, Z., Hauptmann, A. G., & Sebe, N. (2013) · IEEE Transactions on Multimedia · doi:10.1109/tmm.2012.2237023 · Cited by 202 · cited 1×
“The l1-norm penalty in (3) can also be replaced by the l1,2norm as in outlier pursuit [28], [53], [54], [55] and subspace learning [5], [56], [57].”
BackgroundRobust PCA via Outlier Pursuit
Xu, H., Caramanis, C., & Sanghavi, S. (2010) · arXiv (Cornell University) · doi:10.48550/arxiv.1010.4237 · Cited by 192 · cited 1×
“The l1-norm penalty in (3) can also be replaced by the l1,2norm as in outlier pursuit [28], [53], [54], [55] and subspace learning [5], [56], [57].”
BackgroundEriksson, A., & van den Hengel, A. (2010) · 2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2010.5540139 · Cited by 190 · cited 1×
“To induce sparsity, a principled way is to use the convex l1-norm [6], [13], [14], [15], [16], [17], which is the closest convex relaxation of the sparser lp-norm, with compressed sensing being a prominent example.”
MethodAnnihilating Filter-Based Low-Rank Hankel Matrix Approach for Image Inpainting
Jin, K. H., & Ye, J. C. (2015) · IEEE Transactions on Image Processing · doi:10.1109/tip.2015.2446943 · Cited by 177 · cited 3×
“…matrix X ∈ R m × n of rank at most r ≤ d , the following equalities hold: The bilinear spectral penalty in the first equality of (4) has been widely used in low-rank matrix completion and recovery problems, such as RPCA [15], [49], online RPCA [68], matrix completion [69], and image inpainting [25].”
BackgroundMethodInfluentialCompressive principal component pursuit
Wright, J. N., Ganesh, A., Min, K., & Ma, Y. (2013) · Information and Inference A Journal of the IMA · doi:10.1093/imaiai/iat002 · Cited by 177
Fast image deconvolution using closed-form thresholding formulas of regularization
Cao, W., Sun, J. Z., & Xu, Z. (2012) · Journal of Visual Communication and Image Representation · doi:10.1016/j.jvcir.2012.10.006 · Cited by 164
Smoothed Low Rank and Sparse Matrix Recovery by Iteratively Reweighted Least Squares Minimization
Lu, C., Lin, Z., & Yan, S. (2014) · IEEE Transactions on Image Processing · doi:10.1109/tip.2014.2380155 · Cited by 153 · cited 4×
“Compared with the nuclear norm, the Schatten-q norm for 0< q < 1 is equivalent to the lq-norm on singular values and makes a closer approximation to the rank function [32], [33].”
BackgroundMethodInfluentialSelf-citationPractical low-rank matrix approximation under robust L1-norm
Zheng, Y., Liu, G., Sugimoto, S., Yan, S., & Okutomi, M. (2012) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr.2012.6247828 · Cited by 151 · cited 8×
“We also compare our methods with several state-of-the-art methods, such as LMaFit3 [47], RegL14 [16], Unifying [49], factEN5 [15], RPCA6 [44], PSVT7 [45], WNNM8 [46], and LpSq9 [31].”
BackgroundMethodInfluentialRobust Matrix Completion via Joint Schatten p-Norm and lp-Norm Minimization
Nie, F., Wang, H., Cai, X., Huang, H., & Ding, C. (2012) · 2012 IEEE 12th International Conference on Data Mining · doi:10.1109/icdm.2012.160 · Cited by 141 · cited 10×
“We also compare our methods with several state-of-the-art methods, such as LMaFit3 [47], RegL14 [16], Unifying [49], factEN5 [15], RPCA6 [44], PSVT7 [45], WNNM8 [46], and LpSq9 [31].”
BackgroundMethodInfluentialRobust Principal Component Analysis on Graphs
Shahid, N., Kalofolias, V., Bresson, X., Bronstein, M. M., & Vandergheynst, P. (2015) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2015.322 · Cited by 113 · cited 10×
“For instance, it has been observed that the gradient of natural scene images can be better modeled with a heavy-tailed distribution such as hyper-Laplacian distributions ( p ( x ) ∝ e − k | x | α , typically with 0 .”
BackgroundMethodInfluentialRobust Orthonormal Subspace Learning: Efficient Recovery of Corrupted Low-Rank Matrices
Shu, X., Porikli, F., & Ahuja, N. (2014) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr.2014.495 · Cited by 111 · cited 2×
“In [16], [42], [64], [65], [66], the column-orthonormal constraint is imposed on the first factor matrix U .”
BackgroundComplexity analysis of interior point algorithms for non-Lipschitz and nonconvex minimization
Bian, W., Chen, X., & Ye, Y. (2014) · Mathematical Programming · doi:10.1007/s10107-014-0753-5 · Cited by 99 · cited 3×
“In particular, our penalties are Lipschitz, and more tractable and scalable than original Schatten quasi-norm minimization, which is non-Lipschitz and generally NP-hard [32], [39].”
BackgroundMethodEriksson, A. P., & Hengel, A. V. D. (2010) · Cited by 97
Robust Matrix Completion and Corrupted Columns
Chen, Y., Xu, H., Caramanis, C., & Sanghavi, S. (2011) · Cited by 90
Knowledge Adaptation with PartiallyShared Features for Event DetectionUsing Few Exemplars
Ma, Z., Yang, Y., Sebe, N., & Hauptmann, A. G. (2014) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2014.2306419 · Cited by 82
A fast tri-factorization method for low-rank matrix recovery and completion
Liu, Y., Jiao, L., & Shang, F. (2012) · Pattern Recognition · doi:10.1016/j.patcog.2012.07.003 · Cited by 76 · cited 3×
“, L=UV T as in [62], where U ∈ Rm×d, V ∈ Rn×d, and usually d ≪ min(m,n), as well as the matrix tri-factorization (MTF) [48] and factorized data [63] cases.”
BackgroundSelf-citationNonlocal Spectral Prior Model for Low-Level Vision
Wang, S., Zhang, L., & Liang, Y. (2013) · Lecture notes in computer science · doi:10.1007/978-3-642-37431-9_18 · Cited by 71 · cited 1×
“It has been shown in [36] that the singular values of nonlocal matrices in natural images usually exhibit a heavytailed distribution (pðsÞ / e kjsj ), as well as the similar phenomena in natural scenes [37], [38], as shown in Fig.”
BackgroundFast randomized Singular Value Thresholding for Nuclear Norm Minimization
Oh, T.-H., Matsushita, Y., Tai, Y., & Kweon, I. S. (2015) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr.2015.7299078 · Cited by 71 · cited 2×
“While there have been many efforts towards fast SVD computation such as partial SVD [60], the performance of those methods is still unsatisfactory for many real applications [59], [61].”
BackgroundFast singular value thresholding without singular value decomposition
Cai, J., & Osher, S. (2013) · Methods and Applications of Analysis · doi:10.4310/maa.2013.v20.n4.a2 · Cited by 69 · cited 1×
“While there have been many efforts towards fast SVD computation such as partial SVD [60], the performance of those methods is still unsatisfactory for many real applications [59], [61].”
BackgroundMissing Modality Transfer Learning via Latent Low-Rank Constraint
Ding, Z., Shao, M., & Fu, Y. (2015) · IEEE Transactions on Image Processing · doi:10.1109/tip.2015.2462023 · Cited by 65 · cited 1×
“The l1-norm penalty in (3) can also be replaced by the l1,2norm as in outlier pursuit [28], [53], [54], [55] and subspace learning [5], [56], [57].”
BackgroundRobust Principal Component Analysis with Missing Data
Shang, F., Liu, Y., Cheng, J. S.-C., & Cheng, H. (2014) · ACM International Conference on Information and Knowledge Management (CIKM) · doi:10.1145/2661829.2662083 · Cited by 64 · cited 1×
“In [16], [42], [64], [65], [66], the columnorthonormal constraint is imposed on the first factor matrix U .”
BackgroundSelf-citationNuclear-L1 norm joint regression for face reconstruction and recognition with mixed noise
Luo, L., Yang, J. J., Qian, J., & Tai, Y. Y. (2015) · Pattern Recognition · doi:10.1016/j.patcog.2015.06.012 · Cited by 52 · cited 2×
“8 , which correspond to non-convex (cid:96) p - norms) [10], [11], as exhibited by the sparse noise/outliers in low-level vision problems [12] shown in Fig.”
Background
Metadata
- Authors
- Shang, F., Cheng, J., Liu, Y., Luo, Z.-Q., & Lin, Z.
- Venue
- IEEE Transactions on Pattern Analysis and Machine Intelligence
- Published
- 2018-09-01
- Volume
- 40
- Issue
- 9
- Pages
- 2066-2080
- ISSN
- 0162-8828 (ISSN-L)
- 2160-9292
- 1939-3539
- OpenAlex
- W2752077201
- Cited by
- 172
- References
- 74
- FWCI
- 12.73
- Type
- journal article
- Language
- en
- Open access
- green
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.