Semiparametric copula models for biometric score level fusion

In biometric recognition, biometric samples (images of faces, fingerprints, voices, gaits, etc.) of people are compared and matchers (classifiers) indicate the level of similarity between any pair of samples by a score. If we model the joint distribution of all scores by a (semiparametric) Gaussian copula model, the resulting correlation matrix will be structured. It has many zeros and many correlations have a common value. Estimation of these parameters is a problem in constrained semiparametric estimation, a topic that we study in quite some generality in the Statistical Theory part of this thesis. The Biometric Application part of it focuses on score level fusion and models the dependence between classifiers also by semiparametric copula models.
The Statistical part of this thesis studies semiparametric estimation of constrained Euclidean parameters. The restrictions are divided into two cases: the parameter has to be in the image of a continuously differentiable function of a lower dimensional parameter and the parameter has to belong to the zero set of a continuously differentiable function of the parameter.
The Biometric part proposes a semiparametric likelihood ratio-based score level fusion strategy by modelling the marginal individual likelihood ratios nonparametrically and the dependence between them by parametric copulas. The dependence parameter is estimated by pseudo-likelihood estimation and its convergence is discussed. A detailed procedure to train the proposed method is provided and applications on real data for the biometric standard verifications and in forensic scenarios are also demonstrated.