Embedded Biometric Sensor Devices: Design and Implementation on Field Programmable Gate Array

During the research activity in my Ph.D. course, I thoroughly studied the biometric systems and the relevant design and implementation techniques allowing the employment of such systems in embedded devices. I focused my attention on the fingerprint-based recognition and classification systems, and on their implementation on Field Programmable Gate Array (FPGA) devices. I was prompted to study biometric systems mainly because these systems may play a key role in the new emerging market of mobile devices (for example, they are recently available in the new generation of Apple and Samsung smart phones). Such market is rapidly growing and influencing the way people use network resources and functionalities (such as commercial, banking, and government services), requiring a security level higher than in the past. Consequently, novel design techniques and technologies for user recognition and are required to be investigated, in order to provide a secure services and resources access. The traditional authentication systems based on username and password are not able to guarantee a suitable protection level. Unlike password, instead, user biometric information is unique and unchangeable; therefore the biometric identity has the advantage to guarantee that only the authorized users have access to available resources and services. However, traditional biometric approaches involves interactions among a large number of entities: passive access points for user biometric trait acquisition, networked databases for user biometric identity storing, and trusted servers running the user recognition systems. So, traditional systems usually undergo several types of attacks, such as Communication Attack (attacking the channel between the server and the database), Replay Attack (replication of user biometric trait processed during the acquisition phase), and Database Attack (manipulation of the stored user biometric identity). Embedded architectures, instead, provide a more secure and flexible infrastructure, since all elaboration steps are performed on board, so biometric identities are securely managed and stored inside the system without any data leaking out. The goal of this thesis is to illustrate the analysis and results of my research activity focused on the design and development of new fingerprint-based recognition systems for embedded devices. The study of the state-of-the-art about biometric systems led me to realize novel approaches to improve the performance of standard systems in order to enable their employment in embedded devices architectures. Most common literature approaches used to implement fingerprint-based recognition and classification systems are reported to provide a starting-point for understanding the contribution of this work. There are many literature approaches to deal with software systems, but few on design and implementation of embedded hardware prototypes. Referring to the developed and proposed fingerprint-based systems, this thesis represents an advancement of embedded biometrics respect to state-of-the-art. The step-over proposed in this work is focused on: 1. a heuristic fingerprint classification technique, requiring only a little set of images as training dataset; 2. an advanced matching technique for personal recognition based on partial fingerprint, able to enhance the system accuracy; 3. the design and implementation of an efficient fingerprint features extractor; 4. the design and implementation of a quality evaluator of raw fingerprint images (able to identify poor quality areas, such as dry and moist portions), allowing to define a novel flow of image processing steps for user recognition. This thesis is divided into two parts, creating a path connecting the state-of-the-art about biometric systems and the novel implemented approaches. The knowledge of the state-of-the-art about biometrics is fundamental to understand the step over presented in this work. For this reason, in the first part, general characteristics of biometric systems are presented with particular reference to fingerprint-based approaches used in literature to realize embedded systems. The second part proposes the developed innovative sensor. A novel flow of image processing steps for user recognition is outlined. Successively, an efficient micro and macro fingerprint features extractor is illustrated. Then, an advanced matching technique for personal recognition using partial fingerprints is presented. Finally, an innovative fingerprint classification approach based on the fusion of Fuzzy C-Means and Naive-Bayes technique is detailed. Experimental results and comparisons with analogous literature systems show the effectiveness on the proposed sensor. All the innovative approaches proposed in this thesis have been published in international conferences and journals