CONTACTLESS FINGERPRINT BIOMETRICS: ACQUISITION, PROCESSING, AND PRIVACY PROTECTION

Biometrics is defined by the International Organization for Standardization (ISO) as “the automated recognition of individuals based on their behavioral and biological characteristics” Examples of distinctive features evaluated by biometrics, called biometric traits, are behavioral characteristics like the signature, gait, voice, and keystroke, and biological characteristics like the fingerprint, face, iris, retina, hand geometry, palmprint, ear, and DNA. The biometric recognition is the process that permits to establish the identity of a person, and can be performed in two modalities: verification, and identification. The verification modality evaluates if the identity declared by an individual corresponds to the acquired biometric data. Differently, in the identification modality, the recognition application has to determine a person's identity by comparing the acquired biometric data with the information related to a set of individuals. Compared with traditional techniques used to establish the identity of a person, biometrics offers a greater confidence level that the authenticated individual is not impersonated by someone else. Traditional techniques, in fact, are based on surrogate representations of the identity, like tokens, smart cards, and passwords, which can easily be stolen or copied with respect to biometric traits. This characteristic permitted a wide diffusion of biometrics in different scenarios, like physical access control, government applications, forensic applications, logical access control to data, networks, and services. Most of the biometric applications, also called biometric systems, require performing the acquisition process in a highly controlled and cooperative manner. In order to obtain good quality biometric samples, the acquisition procedures of these systems need that the users perform deliberate actions, assume determinate poses, and stay still for a time period. Limitations regarding the applicative scenarios can also be present, for example the necessity of specific light and environmental conditions. Examples of biometric technologies that traditionally require constrained acquisitions are based on the face, iris, fingerprint, and hand characteristics. Traditional face recognition systems need that the users take a neutral pose, and stay still for a time period. Moreover, the acquisitions are based on a frontal camera and performed in controlled light conditions. Iris acquisitions are usually performed at a distance of less than 30 cm from the camera, and require that the user assume a defined pose and stay still watching the camera. Moreover they use near infrared illumination techniques, which can be perceived as dangerous for the health. Fingerprint recognition systems and systems based on the hand characteristics require that the users touch the sensor surface applying a proper and uniform pressure. The contact with the sensor is often perceived as unhygienic and/or associated to a police procedure. This kind of constrained acquisition techniques can drastically reduce the usability and social acceptance of biometric technologies, therefore decreasing the number of possible applicative contexts in which biometric systems could be used. In traditional fingerprint recognition systems, the usability and user acceptance are not the only negative aspects of the used acquisition procedures since the contact of the finger with the sensor platen introduces a security lack due to the release of a latent fingerprint on the touched surface, the presence of dirt on the surface of the finger can reduce the accuracy of the recognition process, and different pressures applied to the sensor platen can introduce non-linear distortions and low-contrast regions in the captured samples. Other crucial aspects that influence the social acceptance of biometric systems are associated to the privacy and the risks related to misuses of biometric information acquired, stored and transmitted by the systems. One of the most important perceived risks is related to the fact that the persons consider the acquisition of biometric traits as an exact permanent filing of their activities and behaviors, and the idea that the biometric systems can guarantee recognition accuracy equal to 100\% is very common. Other perceived risks consist in the use of the collected biometric data for malicious purposes, for tracing all the activities of the individuals, or for operating proscription lists. In order to increase the usability and the social acceptance of biometric systems, researchers are studying less-constrained biometric recognition techniques based on different biometric traits, for example, face recognition systems in surveillance applications, iris recognition techniques based on images captured at a great distance and on the move, and contactless technologies based on the fingerprint and hand characteristics. Other recent studies aim to reduce the real and perceived privacy risks, and consequently increase the social acceptance of biometric technologies. In this context, many studies regard methods that perform the identity comparison in the encrypted domain in order to prevent possible thefts and misuses of biometric data. The objective of this thesis is to research approaches able to increase the usability and social acceptance of biometric systems by performing less-constrained and highly accurate biometric recognitions in a privacy compliant manner. In particular, approaches designed for high security contexts are studied in order improve the existing technologies adopted in border controls, investigative, and governmental applications. Approaches based on low cost hardware configurations are also researched with the aim of increasing the number of possible applicative scenarios of biometric systems. The privacy compliancy is considered as a crucial aspect in all the studied applications. Fingerprint is specifically considered in this thesis, since this biometric trait is characterized by high distinctivity and durability, is the most diffused trait in the literature, and is adopted in a wide range of applicative contexts. The studied contactless biometric systems are based on one or more CCD cameras, can use two-dimensional or three-dimensional samples, and include privacy protection methods. The main goal of these systems is to perform accurate and privacy compliant recognitions in less-constrained applicative contexts with respect to traditional fingerprint biometric systems. Other important goals are the use of a wider fingerprint area with respect to traditional techniques, compatibility with the existing databases, usability, social acceptance, and scalability. The main contribution of this thesis consists in the realization of novel biometric systems based on contactless fingerprint acquisitions. In particular, different techniques for every step of the recognition process based on two-dimensional and three-dimensional samples have been researched. Novel techniques for the privacy protection of fingerprint data have also been designed. The studied approaches are multidisciplinary since their design and realization involved optical acquisition systems, multiple view geometry, image processing, pattern recognition, computational intelligence, statistics, and cryptography. The implemented biometric systems and algorithms have been applied to different biometric datasets describing a heterogeneous set of applicative scenarios. Results proved the feasibility of the studied approaches. In particular, the realized contactless biometric systems have been compared with traditional fingerprint recognition systems, obtaining positive results in terms of accuracy, usability, user acceptability, scalability, and security. Moreover, the developed techniques for the privacy protection of fingerprint biometric systems showed satisfactory performances in terms of security, accuracy, speed, and memory usage.