Characterization of 3D Image-Based Biometric Systems in Dynamic Acquisition Conditions

Abstract

The paper focuses on the analysis of biometric measurements in dynamic acquisition conditions and their impact on the reliability of the recognition judgments. To this aim, a suitable simulator of stereoscopic systems has been designed and realized. It relies on a fully simulated procedure based on the following steps: (i) generation of a set of realistic 3D face models through a proper face simulator software; (ii) definition of an arbitrary trajectory for the face models and stereo images to simulate a set of images acquired in different poses (positions and orientations) of the subject during the movement; (iii) addition of selectable levels of motion blur in a controlled environment, to simulate critical acquisition conditions. This procedure allows ensuring that the recognition results are not due to the natural change of expression of real faces or an imperfect image acquisition device. Moreover, every face model is moved exactly with the same trajectory in front of the stereoscopic system, allowing compare the recognition performances all along the trajectory, also in controlled and under repeatable blur levels. A face biometrics procedure, based on a popular recognition algorithm, is then run on the generated images and the recognition performances are analyzed in detail. The achieved results demonstrated how the motion blur and also the slight differences between the acquired images and the reference ones significantly affect the performance in the recognition of such kinds of systems, thus confirming the usefulness of the proposed simulator.