## A Comparative Evaluation of Fusion Strategies for Multimodal Biometric Verification (2003)

Venue: | In Springer LNCS-2688, 4th Int’l. Conf. Audio- and Video-Based Biometric Person Authentication (AVBPA 2003 |

Citations: | 22 - 6 self |

### BibTeX

@INPROCEEDINGS{Fierrez-aguilar03acomparative,

author = {J. Fierrez-aguilar and J. Ortega-garcia and D. Garcia-romero and J. Gonzalez-rodriguez},

title = {A Comparative Evaluation of Fusion Strategies for Multimodal Biometric Verification},

booktitle = {In Springer LNCS-2688, 4th Int’l. Conf. Audio- and Video-Based Biometric Person Authentication (AVBPA 2003},

year = {2003},

pages = {830--837}

}

### Years of Citing Articles

### OpenURL

### Abstract

Abstract. The aim of this paper, regarding multimodal biometric verification, is twofold: on the one hand, some score fusion strategies reported in the literature are reviewed and, on the other hand, we compare experimentally a selection of them using as monomodal baseline experts: i) our face verification system based on a global face appearance representation scheme, ii) our minutiaebased fingerprint verification system, and iii) our on-line signature verification system based on HMM modeling of temporal functions, on the MCYT multimodal database. A new strategy is also proposed and discussed in order to generate a multimodal combined score by means of Support Vector Machine (SVM) classifiers from which user-independent and user-dependent fusion schemes are derived and evaluated. 1

### Citations

9539 | Statistical Learning Theory - Vapnik - 1998 |

1062 | On Combining Classifiers
- Kittler, Hatef, et al.
- 1998
(Show Context)
Citation Context ...rates for a wide range of decision thresholds with a particular axis scaling. 2 Multimodal Fusion 2.1 Fusion Strategies Biometric multimodality can be studied as a classifier combination problem [2], =-=[9]-=-. Kittler et al. considered in [9] the task of combining classifiers in a probabilistic bayesian framework and provided an example of multimodal biometric verification (fusing speech, frontal and prof... |

478 |
P ttern Recognition
- Theodoridis, Koutroumbas
- 1999
(Show Context)
Citation Context ...referred to as rule-based fusion, because it does not takes into account the actual distribution of outputs from the experts. Multimodal fusion can also be treated as a pattern classification problem =-=[10]-=-. Under this point of view, the scores given by individual expert modalities are considered as input patterns to be labeled as accepted/rejected (for the verification task). Verlinde et al. followed t... |

387 |
Methods of combining multiple classifiers and their applications to handwriting recognition
- Xu, Krzyzak, et al.
- 1992
(Show Context)
Citation Context ... face, for its direct visualness with respect to in situ human interaction; and iii) regarding signature, for its personal, social and legal acceptability as an identification procedure. Some studies =-=[2]-=- have showed that the performance of any single-trait verification system can be improved by unimodal (or monomodal) fusion, i.e., the combination of several verification strategies applied on the sam... |

344 | XM2VTSDB: The extended M2VTS database
- Messer, Matas, et al.
- 1999
(Show Context)
Citation Context ... Description We have randomly selected 50 users from the MCYT Multimodal Database including fingerprint and on-line signature samples [7]. A subset of 50 different users from the XM2VTS face database =-=[14]-=- has been also randomly selected. From both subsets, and thanks to the independence of signature, fingerprint and face traits [3], we have created 50 chimeric individuals comprising fingerprint, signa... |

210 | J.Z.: Information fusion in biometrics
- Ross, Jain, et al.
- 2001
(Show Context)
Citation Context ... applied on the same input data. Even greater verification performance improvement can be expected through the use of multiple biometric characteristics assuming statistical independence between them =-=[3]-=-. Some works related to the multimodal fusion approach are [3]-[6]. In this contribution, and after reviewing some referenced approaches to score fusion in multimodal biometrics, especially those base... |

183 | Person identification using multiple cues
- Brunelli, Falavigna
- 1995
(Show Context)
Citation Context ...Comparative Evaluation of Fusion Strategies 833 defining K ( x , x ) � �( x ), �( x ) as the kernel function and using (2) we obtain: i j i j F � � D( ) sign � y K( , ) b � � * xT � � �i i xi x T � � =-=(4)-=- i�SV � � and b * in its dual form which, together with deciProblem (1) is solved for ( ) 1 l i i sion function (4), avoids manipulating directly the elements of F. The choice for the kernel has been ... |

141 | Integrating faces and fingerprint for personal identification
- Hong, Jain
- 1998
(Show Context)
Citation Context ...ance improvement can be expected through the use of multiple biometric characteristics assuming statistical independence between them [3]. Some works related to the multimodal fusion approach are [3]-=-=[6]-=-. In this contribution, and after reviewing some referenced approaches to score fusion in multimodal biometrics, especially those based on SVM classifiers as they J. Kittler and M.S. Nixon (Eds.): AVB... |

61 |
Evaluation protocol for the extended M2VTS database (XM2VTSDB
- Luttin, Mâıtre
- 1998
(Show Context)
Citation Context ...rol minutiae pattern; ii) Signature: Each signature has been modelled with 6 samples, and iii) Face: Each face has been modelled with 4 samples, according to Configuration II of the Lausanne Protocol =-=[15]-=-. � Testing: i) Targets: 4 more samples of each trait (face, fingerprint and signature) have also been selected for tests (2 from evaluation and 2 from test data of the Lausanne Protocol in the case o... |

49 | Multi-modal identity verification using expert fusion
- Verlinde, Chollet, et al.
- 2000
(Show Context)
Citation Context ...he scores given by individual expert modalities are considered as input patterns to be labeled as accepted/rejected (for the verification task). Verlinde et al. followed this approach and compared in =-=[11]-=- the following pattern classification techniques for multimodal fusion (sorted by relative decreasing performance): Logistic Regression, Maximum a Posteriori, k-Nearest Neighbours classifiers, Multila... |

41 |
Expert conciliation for multimodal person authentication systems using Bayesian statistics
- Bigun, Bigun, et al.
- 1997
(Show Context)
Citation Context ...( ) 1 l i i sion function (4), avoids manipulating directly the elements of F. The choice for the kernel has been in this case a Radial Basis Function (RBF): K 2 2 ( i, j) �exp( � i � j 2 � ) x x x x =-=(5)-=- In [12], the fusion strategy relied on the computation of the decision function D(·). A modification in order to obtain not a final classifier decision, but a combined multimodal score based on the p... |

19 | Multi-modal identity verification using support vector machines (SVM
- Gutschoven, Verlinde
- 2000
(Show Context)
Citation Context ... Kittler and M.S. Nixon (Eds.): AVBPA 2003, LNCS 2688, pp. 830-837, 2003. c Springer-Verlag Berlin Heidelberg 2003A Comparative Evaluation of Fusion Strategies 831 have shown outstanding performance =-=[12]-=-, we will derive fusion schemes, both userindependent and user-dependent, based on the computation of a multimodal combined score. Thereafter, fusion strategies will be compared using our signature, f... |

18 | Complete signal modeling and score normalization for function-based dynamic signature verification
- Ortega-Garcia, Fierrez-Aguilar, et al.
- 2003
(Show Context)
Citation Context ...on a global face appearance representation scheme [16], a minutiae-based fingerprint verification system [17] and an on-line signature verification system based on HMM modelling of temporal functions =-=[18]-=-. 3.3. Multimodal Experimental Procedure For the rule-based fusion approach, all multimodal test scores (200 from clients and 750 from impostors) are used for testing the verification performance. For... |

16 |
The DET Curve in Assessment of Decision Task Performance
- Martin
- 1997
(Show Context)
Citation Context ...ER (equal error rate) which is the point attained when FA and FR rates coincide. Here, the performance of competing systems based on different fusion strategies will be compared by means of DET plots =-=[8]-=-, which are graphical representations of FA vs. FR rates for a wide range of decision thresholds with a particular axis scaling. 2 Multimodal Fusion 2.1 Fusion Strategies Biometric multimodality can b... |

5 |
et al., “MCYT: A Multimodal Biometric Database
- Ortega-Garcia
- 2002
(Show Context)
Citation Context ...ependent and user-dependent, based on the computation of a multimodal combined score. Thereafter, fusion strategies will be compared using our signature, face and fingerprint systems on MCYT database =-=[7]-=-. In verification or authentication (the problem addressed here) a claim is made concerning the identity of a person and the biometric system has to take the binary decision of accepting or rejecting ... |

1 | A Comparative Evaluation of Global Representation-Based Schemes for Face Verification
- Cruz-Llanas, Fierrez-Aguilar, et al.
(Show Context)
Citation Context ...ally used because it makes the comparison of subsequent fusion strategies easier. In particular, we have considered: a face verification system based on a global face appearance representation scheme =-=[16]-=-, a minutiae-based fingerprint verification system [17] and an on-line signature verification system based on HMM modelling of temporal functions [18]. 3.3. Multimodal Experimental Procedure For the r... |

1 |
An Improved Image Enhancement Scheme for Fingerprint Minutiae Extraction in Biometric Identification
- Simon
- 2001
(Show Context)
Citation Context ... fusion strategies easier. In particular, we have considered: a face verification system based on a global face appearance representation scheme [16], a minutiae-based fingerprint verification system =-=[17]-=- and an on-line signature verification system based on HMM modelling of temporal functions [18]. 3.3. Multimodal Experimental Procedure For the rule-based fusion approach, all multimodal test scores (... |

1 |
Fusion strategies
- Fierrez-Aguilar, Ortega-Garcia, et al.
(Show Context)
Citation Context ...le (leading even to error-free combined systems [12], due to the scarceness of data). In the present contribution, a statistical-motivated experimental procedure denominated as Asymptotic Performance =-=[19]-=-, which reduces side effects produced by data scarceness and avoids the uninformative possible “0.0% EER” result, has been used. The proposed statistically-motivated experimental procedure works as fo... |