Revisiting face processing with light field images

Nowadays, in a time where cities contain millions of people and where travelling across the world is becoming easier and easier, the necessity of automatically identifying a person is starting to be compelling. The physical appearance and the behavioural characteristics have been discovered useful to univocally describe a person. The analytic study of the human body measures with the aim of recognising or verifying the identity of a person, is called biometrics, literally “life measure”. In the last century, several biometric traits have been investigated according to the most updated technologies available at the moment, improving recognition, computational time and memory capacity. Starting from the 90?s, research on biometrics has received a huge boost thanks to the interest raised by academic institutions, government agencies and private companies. Moreover, the diffusion of new instruments, able to perform faster analyses, and to store more data, simplifies the development of biometric systems. Together with the advancement of processing machines, innovative acquisition devices providing non conventional data have been developed. The investigation of the impact of new technologies applied to specific topics is a mandatory step in order to improve the performances on biometrics. The main goal of this thesis is to present a non-conventional acquisition technology as light field, to study face analysis performances using images collected with a specific camera, to compare the results with those obtained using data from similar devices and to prove the major benefit provided by the use of up-to-date device over standard cameras in biometric field. When this thesis started, the literature on face analysis with light field data was bare. The scarcity of biometric data (and particularly of human face images) collected with plenoptic cameras has been tackled with a systematic acquisition of a light field face database, now publicly available. Thanks to the collected data, it has been possible to design and develop experiments on face analysis. Moreover, an exhaustive baseline of a comparison between two RGB-D technologies has been created to sustain future studies. During the period of this thesis, the interest on light field technology applied on face analysis has grown and the necessity of a survey on algorithm customized for plenoptic images has become compulsory. Thus, a complete overview on existent methods has been compiled. All the algorithms designed and developed have been tested within the context of the H2020 European Projects with the aim of making faster and more user friendly automatic security border controls.