On Matching Scores for LDA-based Face Verification

We address the problem of face verification using linear discriminant analysis and investigate the issue of matching score 1 . We establish the reason behind the success of the normalised correlation. The improved understanding about the role of metric then naturally leads to a novel way of measuring the distance between a probe image and a model. In extensive experimental studies on the publicly available XM2VTS database 2 using the Lausanne protocol 3 we show that the proposed metric is consistently superior to both the Euclidean distance and normalised correlation matching scores. The effect of various photometric normalisations 4 on the matching scores is also investigated.

[1]  Jiri Matas,et al.  XM2VTSDB: The Extended M2VTS Database , 1999 .

[2]  Jiri Matas,et al.  Effective Implementation of Linear Discriminant Analysis for Face Recognition and Verification , 1999, CAIP.

[3]  Rama Chellappa,et al.  Face recognition using discriminant eigenvectors , 1996, 1996 IEEE International Conference on Acoustics, Speech, and Signal Processing Conference Proceedings.

[4]  Chengjun Liu,et al.  Face Recognition Using Evolutionary Pursuit , 1998, ECCV.

[5]  David J. Kriegman,et al.  Eigenfaces vs. Fisherfaces: Recognition Using Class Specific Linear Projection , 1996, ECCV.

[6]  Chengjun Liu,et al.  Enhanced Fisher linear discriminant models for face recognition , 1998, Proceedings. Fourteenth International Conference on Pattern Recognition (Cat. No.98EX170).

[7]  Alex Pentland,et al.  Probabilistic Visual Learning for Object Representation , 1997, IEEE Trans. Pattern Anal. Mach. Intell..

[8]  Jiri Matas,et al.  Fast face localisation and verification , 1999, Image Vis. Comput..

[9]  Rama Chellappa,et al.  Discriminant analysis of principal components for face recognition , 1998, Proceedings Third IEEE International Conference on Automatic Face and Gesture Recognition.