## Fusing Gabor and LBP Feature Sets for Kernel-Based Face Recognition

### Cached

### Download Links

- [lear.inrialpes.fr]
- [eprints.pascal-network.org]
- [hal.inria.fr]
- [parnec.nuaa.edu.cn]
- DBLP

### Other Repositories/Bibliography

Citations: | 24 - 0 self |

### BibTeX

@MISC{Tan_fusinggabor,

author = {Xiaoyang Tan and Bill Triggs},

title = {Fusing Gabor and LBP Feature Sets for Kernel-Based Face Recognition},

year = {}

}

### OpenURL

### Abstract

Abstract. Extending recognition to uncontrolled situations is a key challenge for practical face recognition systems. Finding efficient and discriminative facial appearance descriptors is crucial for this. Most existing approaches use features of just one type. Here we argue that robust recognition requires several different kinds of appearance information to be taken into account, suggesting the use of heterogeneous feature sets. We show that combining two of the most successful local face representations, Gabor wavelets and Local Binary Patterns (LBP), gives considerably better performance than either alone: they are complimentary in the sense that LBP captures small appearance details while Gabor features encode facial shape over a broader range of scales. Both feature sets are high dimensional so it is beneficial to use PCA to reduce the dimensionality prior to normalization and integration. The Kernel Discriminative Common Vector method is then applied to the combined feature vector to extract discriminant nonlinear features for recognition. The method is evaluated on several challenging face datasets including FRGC 1.0.4, FRGC 2.0.4 and FERET, with promising results. 1

### Citations

1093 | On combining classifiers
- Kittler, Hatef, et al.
- 1998
(Show Context)
Citation Context ...s into a single fused one that is then used in a conventional classifier, whereas decision-level ones combine several classifiers (e.g. based on distinct features) to make a stronger final classifier =-=[11]-=- (this is also called post-classification fusion or mixture of experts). Face recognition is an area that is well-suited to the use (and hence fusion) of multiple classes of descriptors owing to its i... |

1076 |
The use of multiple measurements in taxonomic problems
- Fisher
- 1936
(Show Context)
Citation Context ...features from the combined feature vector z. This is the only point at which class label information is used during training. It is based on a kernelized variant of Linear Discriminant Analysis (LDA) =-=[24]-=- called KDCV [4]. Classical LDA seeks a low-dimensional projection matrix P that maximizes the objective function J(P)= PT SBP P T SWP where SB denotes the between-class and SW the within-class scatte... |

963 | Face recognition: A literature survey
- Zhao, Chellappa, et al.
- 2002
(Show Context)
Citation Context ...ion is finding efficient and discriminative facial appearance descriptors that are resistant to large variations in illumination, pose, facial expression, ageing, partial occlusions and other changes =-=[32]-=-. Most current recognition systems use just one type of features. However for complex tasks such as face recognition, it is often the case that no single feature modality is rich enough to capture all... |

810 | Algorithms for non-negative matrix factorization
- Lee, Seung
- 2001
(Show Context)
Citation Context ...efore fusion. Many dimensionality reduction techniques could be considered – Principal Component Analysis (PCA) [8], Independent Component Analysis (ICA) [20], Non-negative Matrix Factorization (NMF) =-=[14]-=-, and Canonical Correlation Analysis (CCA) [13] to mention only some of the linear ones – but here we find that simple PCA suffices. The reduced feature vectors are separately normalized before being ... |

805 | The FERET Evaluation Methodology for Face-Recognition Algorithms
- Phillips, Moon, et al.
- 2000
(Show Context)
Citation Context ...s containing large lighting variations similar to those encountered in natural images taken under uncontrolled conditions: the Face Recognition Grand Challenge 1.0.4 and 2.0.4 datasets [21] and FERET =-=[22]-=-. 2 Related Work Information fusion for visual recognition can occur at feature-level or at decision-level [10]. Feature-level methods combine several incoming feature sets into a single fused one tha... |

646 |
Learning with kernels: Support vector machines, regularization, optimization, and beyond
- Schölkopf, Smola
- 2002
(Show Context)
Citation Context ...bined representation. Finally, we use a kernel discriminant to extract as much information as possible from the resulting combined features. Methods such as Kernel Principal Component Analysis (KPCA) =-=[25]-=- have proven to be effective nonlinear feature extractors and here we use a related discriminative method, Kernel Discriminative Common Vectors (KDCV). Like other kernel methods, KDCV uses a nonlinear... |

618 | Multiresolution gray-scale and rotation invariant texture classification with local binary patterns
- Ojala, Pietikinen, et al.
- 2002
(Show Context)
Citation Context ...we show that face recognition performance can be significantly improved by combining two of the most successful local appearance descriptors, Gabor wavelets [12,28,15] and Local Binary Patterns (LBP) =-=[18,19,2]-=-. LBP is basically a finescale descriptor that captures small texture details. Local spatial invariance is achieved by locally pooling (histogramming) the resulting texture codes. Given that it is als... |

589 |
Analysis of a complex of statistical variables into principal components
- Hotelling
- 1933
(Show Context)
Citation Context ...onality’ problems. To counteract this we run dimensionality reduction on each modality before fusion. Many dimensionality reduction techniques could be considered – Principal Component Analysis (PCA) =-=[8]-=-, Independent Component Analysis (ICA) [20], Non-negative Matrix Factorization (NMF) [14], and Canonical Correlation Analysis (CCA) [13] to mention only some of the linear ones – but here we find that... |

534 | Distortion invariant object recognition in the dynamic link architecture
- Lades, Vorbruggen, et al.
- 1993
(Show Context)
Citation Context ...] and face recognition[16]. In this paper, we show that face recognition performance can be significantly improved by combining two of the most successful local appearance descriptors, Gabor wavelets =-=[12,28,15]-=- and Local Binary Patterns (LBP) [18,19,2]. LBP is basically a finescale descriptor that captures small texture details. Local spatial invariance is achieved by locally pooling (histogramming) the res... |

478 |
von der Malsburg. Face Recognition by Elastic
- Wiskott, Fellous, et al.
- 1997
(Show Context)
Citation Context ...] and face recognition[16]. In this paper, we show that face recognition performance can be significantly improved by combining two of the most successful local appearance descriptors, Gabor wavelets =-=[12,28,15]-=- and Local Binary Patterns (LBP) [18,19,2]. LBP is basically a finescale descriptor that captures small texture details. Local spatial invariance is achieved by locally pooling (histogramming) the res... |

443 |
Learning with Kernels: Support Vector
- Scholkopf, Smola
- 2002
(Show Context)
Citation Context ...bined representation. Finally, we use a kernel discriminant to extract as much information as possible from the resulting combined features. Methods such as Kernel Principal Component Analysis (KPCA) =-=[25]-=- have proven to be effective nonlinear feature extractors and here we use a related discriminative method, Kernel Discriminative Common Vectors (KDCV). Like other kernel methods, KDCV uses a nonlinear... |

429 | Using Discriminant Eigenfeatures for Image Retrieval
- Swets, Weng
- 1996
(Show Context)
Citation Context ... along these directions, LDA considers the corresponding classes to be perfectly separable. Techniques proposed to solve this classical problem include the perturbation method [31], two stage PCA+LDA =-=[26]-=-, and the null space methods pioneered by Chen et al. [7]. The latter have dominated research in recent years. They focus only on the null space of SW ,sotheyarereally complements to traditional LDA n... |

386 |
A comparative study of texture measures with classification based on feature distributions. 29(1):51–59
- Ojala, Pietikainen, et al.
- 1996
(Show Context)
Citation Context ...we show that face recognition performance can be significantly improved by combining two of the most successful local appearance descriptors, Gabor wavelets [12,28,15] and Local Binary Patterns (LBP) =-=[18,19,2]-=-. LBP is basically a finescale descriptor that captures small texture details. Local spatial invariance is achieved by locally pooling (histogramming) the resulting texture codes. Given that it is als... |

290 | Overview of the face recognition grand challenge
- Phillips, Flynn, et al.
- 2005
(Show Context)
Citation Context ...gnition datasets containing large lighting variations similar to those encountered in natural images taken under uncontrolled conditions: the Face Recognition Grand Challenge 1.0.4 and 2.0.4 datasets =-=[21]-=- and FERET [22]. 2 Related Work Information fusion for visual recognition can occur at feature-level or at decision-level [10]. Feature-level methods combine several incoming feature sets into a singl... |

222 | Face description with local binary patterns: application to face recognition
- Ahonen, Hadid, et al.
- 2006
(Show Context)
Citation Context ...we show that face recognition performance can be significantly improved by combining two of the most successful local appearance descriptors, Gabor wavelets [12,28,15] and Local Binary Patterns (LBP) =-=[18,19,2]-=-. LBP is basically a finescale descriptor that captures small texture details. Local spatial invariance is achieved by locally pooling (histogramming) the resulting texture codes. Given that it is als... |

220 | Information fusion in biometrics
- Ross, Jain
- 2003
(Show Context)
Citation Context ...Selecting appropriate and complementary component features is crucial for good performance. There is some work on fusing different biometric modalities (e.g. face and speech [3], face and fingerprint =-=[23]-=-), but most studies concentrate on fusing different representations of a single underlying modality (e.g. 2D and 3D facial shape in [6]). Our work belongs to this category, studying the effectiveness ... |

193 | Discriminant Analysis of Principal Components for Face Recognition
- Zhao, Chellappa, et al.
- 1998
(Show Context)
Citation Context ... a nontrivial projection along these directions, LDA considers the corresponding classes to be perfectly separable. Techniques proposed to solve this classical problem include the perturbation method =-=[31]-=-, two stage PCA+LDA [26], and the null space methods pioneered by Chen et al. [7]. The latter have dominated research in recent years. They focus only on the null space of SW ,sotheyarereally compleme... |

168 |
A new LDA-based face recognition system which can solve the small sample size problem
- Chen, Liao, et al.
- 2000
(Show Context)
Citation Context ...lasses to be perfectly separable. Techniques proposed to solve this classical problem include the perturbation method [31], two stage PCA+LDA [26], and the null space methods pioneered by Chen et al. =-=[7]-=-. The latter have dominated research in recent years. They focus only on the null space of SW ,sotheyarereally complements to traditional LDA not stabilized variants of it. They optimize the null spac... |

165 | Face recognition with local binary patterns, in: ECCV, 2004, pp. 469–481. Please cite this article as: L. Pishchulin, et al., Image warping for face recognition: From local optimality towards global optimization
- Ahonen, Hadid, et al.
(Show Context)
Citation Context ... monotonic gray-level transformations, and they have been shown to have high discriminative power for texture classification [18]. T. Ahonen et al. introduced an LBP based method for face recognition =-=[1]-=- that divides the face into a regular grid of cells and histograms the uniform LBP’s within each cell. Finally, the cell-level histograms are concatenated to produce a global descriptor vector. Like t... |

143 |
Independent component analysis—a new concept? Signal Process
- Comon
- 1994
(Show Context)
Citation Context ...un dimensionality reduction on each modality before fusion. Many dimensionality reduction techniques could be considered – Principal Component Analysis (PCA) [8], Independent Component Analysis (ICA) =-=[20]-=-, Non-negative Matrix Factorization (NMF) [14], and Canonical Correlation Analysis (CCA) [13] to mention only some of the linear ones – but here we find that simple PCA suffices. The reduced feature v... |

139 |
A method of combining multiple experts for the recognition of unconstrained handwritten numerals, IEEE Transaction of Pattern Analysis and Machine Intelligence
- Suen
- 1995
(Show Context)
Citation Context ...ombining complementary feature sets has thus become an active research topic in pattern recognition, with successful applications in many challenging tasks including handwritten character recognition =-=[9]-=- and face recognition[16]. In this paper, we show that face recognition performance can be significantly improved by combining two of the most successful local appearance descriptors, Gabor wavelets [... |

138 |
Score normalization in multimodal biometric systems
- Jain, Nandakumar, et al.
- 2005
(Show Context)
Citation Context ...on. Here we evaluate and normalize the two modalities independently before combining them (although some works argue that it can be more effective to fuse modalities at an earlier stage of processing =-=[10]-=-). Both feature sets are high-dimensional (typically at least 10 4 -D) and simply concatenating them would tend to exacerbate any ‘curse of dimensionality’ problems. To counteract this we run dimensio... |

113 | Enhanced local texture feature sets for face recognition under difficult lighting conditions
- Tan, Triggs
- 2007
(Show Context)
Citation Context ...range of output variations; and sigmoid-function based compression of any remaining signal highlights. A detailed description of this simple but very effective normalization procedure can be found in =-=[27]-=-. Some examples of preprocessed images are shown in Fig. 3. The downsampling factor for the Gabor features is set to 64, resulting a dimensionality of 10 240 (128 · 128 · 40/64), while the cell size o... |

97 | Fusion of face and speech data for person identity verification
- Ben-Yacoub, Abdeljaoued, et al.
- 1999
(Show Context)
Citation Context ...on methods are different. Selecting appropriate and complementary component features is crucial for good performance. There is some work on fusing different biometric modalities (e.g. face and speech =-=[3]-=-, face and fingerprint [23]), but most studies concentrate on fusing different representations of a single underlying modality (e.g. 2D and 3D facial shape in [6]). Our work belongs to this category, ... |

73 | Local Gabor Binary Pattern Histogram Sequence (LGBPHS): A novel non-statistical model for face representation and recognition
- Zhang, Shan, et al.
- 2005
(Show Context)
Citation Context ...ontrast, Gabor features [12,28,15] encode facial shape and appearance information over a range of coarser scales (although they have also been used as a preprocessing stage for LBP feature extraction =-=[30]-=-). Both representations are S.K. Zhou et al. (Eds.): AMFG 2007, LNCS 4778, pp. 235–249, 2007. c○ Springer-Verlag Berlin Heidelberg 2007s236 X. Tan and B. Triggs rich in information and computationally... |

68 | Discriminative common vectors for face recognition - Cevikalp, Neamtu, et al. - 2005 |

62 |
Kernel and nonlinear canonical correlation analysis
- Lai, Fyfe
- 2000
(Show Context)
Citation Context ...hniques could be considered – Principal Component Analysis (PCA) [8], Independent Component Analysis (ICA) [20], Non-negative Matrix Factorization (NMF) [14], and Canonical Correlation Analysis (CCA) =-=[13]-=- to mention only some of the linear ones – but here we find that simple PCA suffices. The reduced feature vectors are separately normalized before being concatenated into a single combined feature vec... |

40 | Capitalize on dimensionality increasing techniques for improving face recognition grand challenge performance
- Liu
- 2006
(Show Context)
Citation Context ...] and face recognition[16]. In this paper, we show that face recognition performance can be significantly improved by combining two of the most successful local appearance descriptors, Gabor wavelets =-=[12,28,15]-=- and Local Binary Patterns (LBP) [18,19,2]. LBP is basically a finescale descriptor that captures small texture details. Local spatial invariance is achieved by locally pooling (histogramming) the res... |

34 | A Shape- and TextureBased Enhanced Fisher Classifier for Face Recognition
- Liu, Wechsler
- 2001
(Show Context)
Citation Context ...eature sets has thus become an active research topic in pattern recognition, with successful applications in many challenging tasks including handwritten character recognition [9] and face recognition=-=[16]-=-. In this paper, we show that face recognition performance can be significantly improved by combining two of the most successful local appearance descriptors, Gabor wavelets [12,28,15] and Local Binar... |

34 | Combining classifiers for face recognition
- Lu, Wang, et al.
- 2003
(Show Context)
Citation Context ... the use (and hence fusion) of multiple classes of descriptors owing to its inherent complexity and need for fine distinctions. Much of the past work in this area adopts classifier-level fusion, e.g. =-=[6,17]-=-.For example, in [17] PCA, ICA and LDA provide the component subspaces for classifier combination. Each test sample is separately projected into these three subspaces and the resulting distance matric... |

13 |
Discriminative common vector method with kernels
- Cevikalp, Neamtu, et al.
- 2006
(Show Context)
Citation Context ...that simple PCA suffices. The reduced feature vectors are separately normalized before being concatenated into a single combined feature vector. Finally the Kernel Discriminative Common Vector (KDCV) =-=[4]-=- method is applied to the combined feature vector to provide effective multi-class recognition from relatively few training examples. To illustrate the effectiveness of our approach we present experim... |

13 |
Multi-biometrics using facial appearance, shape and temperature
- Chang, Bowyer, et al.
- 2004
(Show Context)
Citation Context ... the use (and hence fusion) of multiple classes of descriptors owing to its inherent complexity and need for fine distinctions. Much of the past work in this area adopts classifier-level fusion, e.g. =-=[6,17]-=-.For example, in [17] PCA, ICA and LDA provide the component subspaces for classifier combination. Each test sample is separately projected into these three subspaces and the resulting distance matric... |

6 |
M.: Face description with local binary patterns: Application to face recognition
- Ahonen, Hadid, et al.
- 2006
(Show Context)
Citation Context |

1 |
Generalized k-l transform based combined feature extraction
- Yang, Yang
- 2002
(Show Context)
Citation Context ...tivated by the belief that the original features are a richer representation than distance matrices or individual classifier decisions, several works have studied feature-level fusion. J. Yang et al. =-=[29]-=- concatenate different features into a single vector and use Generalized PCA for feature exaction. C. Liu et al. [16] concatenate shape and texture information in a reduced subspace then use an enhanc... |

1 | fb fc dup1 dup2 Fisherfaces 94 73 55 31 Ahonen – LBP 96 82 59 52 - Method - 2007 |