## Training Support Vector Machines: an Application to Face Detection (1997)

Citations: | 565 - 1 self |

### BibTeX

@INPROCEEDINGS{Osuna97trainingsupport,

author = {Edgar Osuna and Robert Freund and Federico Girosi},

title = {Training Support Vector Machines: an Application to Face Detection},

booktitle = {},

year = {1997},

pages = {130--136}

}

### Years of Citing Articles

### OpenURL

### Abstract

We investigate the application of Support Vector Machines (SVMs) in computer vision. SVM is a learning technique developed by V. Vapnik and his team (AT&T Bell Labs.) that can be seen as a new method for training polynomial, neural network, or Radial Basis Functions classifiers. The decision surfaces are found by solving a linearly constrained quadratic programming problem. This optimization problem is challenging because the quadratic form is completely dense and the memory requirements grow with the square of the number of data points. We present a decomposition algorithm that guarantees global optimality, and can be used to train SVM's over very large data sets. The main idea behind the decomposition is the iterative solution of sub-problems and the evaluation of optimality conditions which are used both to generate improved iterative values, and also establish the stopping criteria for the algorithm. We present experimental results of our implementation of SVM, and demonstrate the ...

### Citations

8984 | Statistical Learning Theory
- Vapnik
- 1998
(Show Context)
Citation Context ...sional spaces (10 2 \Gamma 10 3 ). In this paper we concentrate on the Support Vector Machine (SVM), a pattern classification algorithm recently developed by V. Vapnik and his team at AT&T Bell Labs. =-=[1, 3, 4, 12]-=-. SVM can be seen as a new way to train polynomial, neural network, or Radial Basis Functions classifiers. While most of the techniques used to train the above mentioned classifiers are based on the i... |

2173 | Support-vector networks
- Cortes, Vapnik
- 1995
(Show Context)
Citation Context ...sional spaces (10 2 \Gamma 10 3 ). In this paper we concentrate on the Support Vector Machine (SVM), a pattern classification algorithm recently developed by V. Vapnik and his team at AT&T Bell Labs. =-=[1, 3, 4, 12]-=-. SVM can be seen as a new way to train polynomial, neural network, or Radial Basis Functions classifiers. While most of the techniques used to train the above mentioned classifiers are based on the i... |

1292 | A training algorithm for optimal margin classifiers
- Boser, Guyon, et al.
- 1992
(Show Context)
Citation Context ...sional spaces (10 2 \Gamma 10 3 ). In this paper we concentrate on the Support Vector Machine (SVM), a pattern classification algorithm recently developed by V. Vapnik and his team at AT&T Bell Labs. =-=[1, 3, 4, 12]-=-. SVM can be seen as a new way to train polynomial, neural network, or Radial Basis Functions classifiers. While most of the techniques used to train the above mentioned classifiers are based on the i... |

582 | Example-based learning for view-based human face detection
- Sung, Poggio
- 1998
(Show Context)
Citation Context ...Networks [2, 9, 11], detection of face features and use of geometrical constraints [13], density estimation of the training data [6], labeled graphs [5] and clustering and distribution-based modeling =-=[10]-=-. Out of all these previous works, the results of Sung and Poggio [10], and Rowley et al. [9] reflect systems with very high detection rates and low false positive rates. Sung and Poggio use clusterin... |

267 |
and B.Sz.-Nagy, â€œFunctional Analysis
- Riesz
- 1990
(Show Context)
Citation Context ...ace becomes particularly simple because: z T (x)z(y) = 1 X i=1 ff i / i (x)/ i (y) = K(x; y) where the last equality comes from the MercerHilbert -Schmidt theorem for positive definite functions (see =-=[8]-=-, pp. 242--246). The QP problem that has to be solved now is exactly the same as in eq. (1), with the exception that the matrix D has now elements D ij = y i y j K(x i ; x j ). As a result of this cho... |

208 | Probabilistic visual learning for object detection
- Moghaddam, Pentland
- 1995
(Show Context)
Citation Context ...erent techniques in the last few years. This techniques include Neural Networks [2, 9, 11], detection of face features and use of geometrical constraints [13], density estimation of the training data =-=[6]-=-, labeled graphs [5] and clustering and distribution-based modeling [10]. Out of all these previous works, the results of Sung and Poggio [10], and Rowley et al. [9] reflect systems with very high det... |

158 | Human face detection in visual scenes
- Rowley, Baluja, et al.
- 1995
(Show Context)
Citation Context ... the appropriate class (face/non-face). 3.1 Previous Systems The problem of face detection has been approached with different techniques in the last few years. This techniques include Neural Networks =-=[2, 9, 11]-=-, detection of face features and use of geometrical constraints [13], density estimation of the training data [6], labeled graphs [5] and clustering and distribution-based modeling [10]. Out of all th... |

145 | Simplified support vector decision rules
- Burges
- 1996
(Show Context)
Citation Context |

111 |
Human face detection in a complex background
- Yang, Huang
- 1994
(Show Context)
Citation Context ...f face detection has been approached with different techniques in the last few years. This techniques include Neural Networks [2, 9, 11], detection of face features and use of geometrical constraints =-=[13]-=-, density estimation of the training data [6], labeled graphs [5] and clustering and distribution-based modeling [10]. Out of all these previous works, the results of Sung and Poggio [10], and Rowley ... |

74 | Large-Scale Linearly Constrained Optimization
- Murtagh, Saunders
- 1978
(Show Context)
Citation Context ...of the decision surface has changed. 2.4 Implementation and Results We have implemented the decomposition algorithm using MINOS 5.4 as the solver of the sub-problems. For information on MINOS 5.4 see =-=[7]-=-. The computational results that we present in this section have been obtained using real data from our Face Detection System, which is described in Section 3. Figures 3a and 3b show the training time... |

34 |
Original approach for the localisation of objects in images
- Vaillant, Monrocq, et al.
- 1994
(Show Context)
Citation Context ... the appropriate class (face/non-face). 3.1 Previous Systems The problem of face detection has been approached with different techniques in the last few years. This techniques include Neural Networks =-=[2, 9, 11]-=-, detection of face features and use of geometrical constraints [13], density estimation of the training data [6], labeled graphs [5] and clustering and distribution-based modeling [10]. Out of all th... |

20 | Determination of face position and pose with learned representation based on labled graphs
- Kruger, Potzsch, et al.
- 1996
(Show Context)
Citation Context ...the last few years. This techniques include Neural Networks [2, 9, 11], detection of face features and use of geometrical constraints [13], density estimation of the training data [6], labeled graphs =-=[5]-=- and clustering and distribution-based modeling [10]. Out of all these previous works, the results of Sung and Poggio [10], and Rowley et al. [9] reflect systems with very high detection rates and low... |

18 |
Detection and localization of faces on digital images
- Burel, Carel
- 1994
(Show Context)
Citation Context ... the appropriate class (face/non-face). 3.1 Previous Systems The problem of face detection has been approached with different techniques in the last few years. This techniques include Neural Networks =-=[2, 9, 11]-=-, detection of face features and use of geometrical constraints [13], density estimation of the training data [6], labeled graphs [5] and clustering and distribution-based modeling [10]. Out of all th... |