#### DMCA

## Gool, L.: Weighted collaborative representation and classification of images

Venue: | In: ICPR. (2012 |

Citations: | 4 - 4 self |

### Citations

964 | Regularization and variable selection via the elastic net - Zou, Hastie - 2005 |

933 | Robust face recognition via sparse representation - Wright, Yang, et al. - 2009 |

548 |
Statistical Analysis With Missing Data, 2nd edition
- Little, Rubin
- 2002
(Show Context)
Citation Context ...(XTΩ−1/2X)−1XTΩ−1/2y (8) When Ω is a diagonal matrix, GLS boils down to Weighted Least Squares (WLS). When Ω is not directly known, it can be estimated as in Feasible Generalized Least Squares (FGLS) =-=[2]-=-. First, use OLS and obtain the residuals u, and take for Ω the diagonal matrix of squared residuals: ûOLS = y −Xβ̂OLS , ΩOLS = diag(ûOLS)2 (9) Then β̂FGLS1 is estimated: β̂FGLS1 = (X TΩ−1/2OLSX) −1... |

510 |
The AR face database
- Martinez, Benavente
- 1998
(Show Context)
Citation Context ...CR, by empirically setting κ2 as the mean value of XTΩ−1/2WCRX , and κ1 to 0.1 from this value, respectively. 5 Experimental results 5.1 Benchmark setup In our experiments we use the AR face database =-=[3]-=- with the same settings as in [10, 7]. There are 100 individuals for a total of 700 training and 700 testing face images of size 60×43. Complementary experiments are conducted on GTSRB traffic signs d... |

441 | Efficient sparse coding algorithms
- Lee, Battle, et al.
(Show Context)
Citation Context ...s much larger with its 43 classes, 39209 training and 12630 testing images. All the features are l2 normalized before and after projections in all experiments. SRC uses either the Feature Sign (FeSg) =-=[1]-=-, Homotopy (Hmtp) or L1LS algorithm for solving the l1 minimization. 5.2 l1, l2 and data dimensionality First, we study the role of l1 and l2 regularization versus the dimensionality of the data and r... |

375 |
A generalized inverse for matrices
- Penrose
- 1955
(Show Context)
Citation Context ...ch runs a query-dependent optimization. When the number of data samples (X) exceeds data dimensionality, the computation of P can be troublesome. A solution comes from the Moore-Penrose pseudoinverse =-=[4]-=-: one can work on the transposed data in order to compute the pseudoinverse P = ((XXT + λCRI)−1X)T (20) Adapting the computation of P using eq. (20) or eq. (18), allows CRC to scale well with either v... |

107 | Sparse representation or collaborative representation: Which helps face recognition
- Zhang, Yang, et al.
(Show Context)
Citation Context ... solution of the fitting problem is desirable, which is the case for the least squares formulation. When all samples contribute to the solution, the method is coined Collaborative Representation (CR) =-=[10]-=-. Aside from the basic least squares criterion for best fitting new observations, other constraints are considered in the literature as well. For stabilizing the coefficients of the least squares deco... |

34 | The German Traffic Sign Recognition Benchmark: A multi-class classification competition
- Stallkamp, Schlipsing, et al.
- 2011
(Show Context)
Citation Context ...e same settings as in [10, 7]. There are 100 individuals for a total of 700 training and 700 testing face images of size 60×43. Complementary experiments are conducted on GTSRB traffic signs database =-=[5]-=-. This is much larger with its 43 classes, 39209 training and 12630 testing images. All the features are l2 normalized before and after projections in all experiments. SRC uses either the Feature Sign... |

20 | Relaxed collaborative representation for pattern classification
- Yang, Zhang, et al.
(Show Context)
Citation Context ...ions, improvement via channel weighting (when k2 = 0) takes over, but is small. WCRC reaches 96%, 2% better than the best CRC results. The WCRC result on AR is similar to the one recently reported by =-=[9]-=- for the Relaxed Collaborative Representation (RCR) Classifier (RCRC), 96.0%- WCRC vs. 95.9%-RCRC. Instead of moving from CR, eq. (4), to a WCR weighted formulation as we do in 3ΓWCR, κ1, κ2 are corre... |

12 |
Gool, Sparse representation based projections
- Timofte, Van
(Show Context)
Citation Context ...̂ = β̂SR, then the resulting decision is the Sparse Representation-based Classifier (SRC) decision. Another, faster, approach is to directly use the weights in absolute values as deciding information =-=[6]-=-. Thus, wc(y) = ‖β̂c‖1, class(y) = arg max c wc (17) If in eq. (17) β̂ = β̂SR, then the resulting decision is the SRCw decision based on coefficients. For a Collaborative Representation Classifier wit... |

9 | Gool. Iterative nearest neighbors for classification and dimensionality reduction
- Timofte, Van
(Show Context)
Citation Context ...the mean value of XTΩ−1/2WCRX , and κ1 to 0.1 from this value, respectively. 5 Experimental results 5.1 Benchmark setup In our experiments we use the AR face database [3] with the same settings as in =-=[10, 7]-=-. There are 100 individuals for a total of 700 training and 700 testing face images of size 60×43. Complementary experiments are conducted on GTSRB traffic signs database [5]. This is much larger with... |