A structured approach for the metrological characterization of biometric systems based on face recognition

The security systems based on the face recognition through video surveillance systems are of great interest. Their use is important in several areas including airport security, identification of individuals, access control to critical areas and so on. These systems are based either on the measurement of parameters details of a human face or on a global approach considering the face as a whole. While in the first case values are associated to geometrical parameters on the face, in the second they are produced by the projection of the face images in a face space determined for example with a Principal or Independent Component Analysis. The recognition is then performed by comparison of the values obtained with those of a set of parameters stored in a database, or measurable by a reference image. The result of the comparison between the values measured and stored in the database is not deterministic because the two sets of values are affected by random type uncertainties. In these circumstances the decision on recognition of a face is certainly subject to a risk of being wrong just because of measurement uncertainty. Target of this paper is the proposal of a new approach for the quantitative assessment of the measurement uncertainty and consequently the risk in decision-making, based on the definition and identification of a model for the evaluation of the uncertainty in face related biometric measurements.