Hardware/software architectures for iris biometrics

• Autores: Judith Liu Jiménez
• Directores de la Tesis: Raúl Sánchez Reillo (dir. tes.)
• Lectura: En la Universidad Carlos III de Madrid ( España ) en 2010
• Idioma: inglés
• Tribunal Calificador de la Tesis: Andrzej Pacut (presid.), Luis Mengibar Pozo (secret.), Marcos Faúndez Zanuy (voc.), Mariano López García (voc.), Bernadette Dorizzi (voc.)
• Materias:
  • Ciencias de la vida
    • Biometría
• Enlaces
  • Tesis en acceso abierto en: e-Archivo
• Resumen

  • Nowadays, the necessity of identifying users of facilities and services has become quite important not only to determine who accesses a system and/or service, but also to determine which privileges should be provided to each user. For achieving such identification, Biometrics is emerging as a technology that provides a high level of security, as well as being convenient and comfortable for the citizen. Most biometric systems are based on computer solutions, where the identification process is performed by servers or workstations, whose cost and processing time make them not feasible for some situations. However, Microelectronics can provide a suitable solution without the need of complex and expensive computer systems. Microelectronics is a subfield of Electronics and as the name suggests, is related to the study, development and/or manufacturing of electronic components, i.e. integrated circuits (ICs). We have focused our research in a concrete field of Microelectronics: hardware/software co-design. This technique is widely used for developing specific and high computational cost devices. Its basis relies on using both hardware and software solutions in an effective way, thus, obtaining a device faster than just a software solution, or smaller devices that use dedicated hardware developed for all the processes. The questions on how we can obtain an effective solution for Biometrics will be solved considering all the different aspects of these systems. In this Thesis, we have made two important contributions: the first one for a verification system based on ID token and secondly, a search engine used for massive recognition systems, both of them related to Iris Biometrics. The first relevant contribution is a biometric system architecture proposal based on ID tokens in a distributed system. In this contribution, we have specified some considerations to be done in the system and describe the different functionalities of the elements which form it, such as the central servers and/or the terminals. The main functionality of the terminal is just left to acquiring the initial biometric raw data, which will be transmitted under security cryptographic methods to the token, where all the biometric process will be performed. The ID token architecture is based on Hardware/software co-design. The architecture proposed, independent of the modality, divides the biometric process into hardware and software in order to achieve further performance functions, more than in the existing tokens. This partition considers not only the decrease of computational time hardware can provide, but also the reduction of area and power consumption, the increase in security levels and the effects on performance in all the design. To prove the proposal made, we have implemented an ID token based on Iris Biometrics following our premises. We have developed different modules for an iris algorithm both in hardware and software platforms to obtain results necessary for an effective combination of same. We have also studied different alternatives for solving the partition problem in the Hardware/software co-design issue, leading to results which point out tabu search as the fastest algorithm for this purpose. Finally, with all the data obtained, we have been able to obtain different architectures according to different constraints. We have presented architectures where the time is a major requirement, and we have obtained 30% less processing time than in all software solutions. Likewise, another solution has been proposed which provides less area and power consumption. When considering the performance as the most important constraint, two architectures have been presented, one which also tries to minimize the processing time and another which reduces hardware area and power consumption. In regard the security we have also shown two architectures considering time and hardware area as secondary requirements. Finally, we have presented an ultimate architecture where all these factors were considered. These architectures have allowed us to study how hardware improves the security against authentication attacks, how the performance is influenced by the lack of floating point operations in hardware modules, how hardware reduces time with software reducing the hardware area and the power consumption. The other singular contribution made is the development of a search engine for massive identification schemes, where time is a major constraint as the comparison should be performed over millions of users. We have initially proposed two implementations: following a centralized architecture, where memories are connected to the microprocessor, although the comparison is performed by a dedicated hardware co-processor, and a second approach, where we have connected the memory driver directly in the hardware coprocessor. This last architecture has showed us the importance of a correct connection between the elements used when time is a major requirement. A graphical representation of the different aspects covered in this Thesis is presented in Fig.1, where the relation between the different topics studied can be seen. The main topics, Biometrics and Hardware/Software Co-design have been studied, where several aspects of them have been described, such as the different Biometric modalities, where we have focussed on Iris Biometrics and the security related to these systems. Hardware/Software Co-design has been studied by presenting different design alternatives and by identifying the most suitable configuration for ID Tokens. All the data obtained from this analysis has allowed us to offer two main proposals: The first focuses on the development of a fast search engine device, and the second combines all the factors related to both sciences with regards ID tokens, where different aspects have been combined in its Hardware/Software Design. Both approaches have been implemented to show the feasibility of our proposal. Finally, as a result of the investigation performed and presented in this thesis, further work and conclusions can be presented as a consequence of the work developed.-----------------------------------------------------------------------------------------