Identificador: TDX:1468
Autors: Cañellas Alberich, Nicolau
Resum:
Fingerprint-based biometrics is one of the more reliable identification methods, and a serious candidate for being used in the daily life. In recent years a lot of new devices incorporate fingerprint biometrics and it is not strange the utilization of automatic fingerprint identification systems for monitoring the access into restricted areas.<br/>The society is evolving towards a new kind of smart cards, joining a fingerprint sensor together with a device capable of performing all of the biometric identification steps. In this framework, the thesis focuses in the integration of biometric systems and smart cards; the target is the implementation of an embedded security system, based in fingerprint biometrics, in order to avoid fraudulent accesses by means of identity verification.<br/>Traditionally, the algorithms used in fingerprint features extraction have been based in the recursive iteration of complex image processing functions. These algorithms have been designed looking only for the correct feature extraction but, until now, there is not any algorithm designed bearing in mind a cost or a portability optimization; the systems have been developed over a personal computer based platform, or using a high feature (and cost) microprocessor, or over an specific digital signal processing (DSP) device.<br/>This work develops a new algorithm for the extraction of the fingerprint physical details (minutiae) directly from a grey scale image; the algorithm does not need any product or division and neither any floating point operation. As the correct estimation of the ridge lines direction usually becomes the most critical step, and computationally most expensive, of the minutiae extraction algorithms, it has also been developed a specific algorithm for this specific task.<br/>In order to develop an real-time automatic identification system, fitted to be implemented in low cost microprocessors, it has been carried out the co-design of a hardware - software system. So, the respective coprocessors have been designed: the one related to the hardware implementation of the ridge lines directions estimation and other dedicated to the rest of critical tasks; these have been identified executing the software version of the algorithm and analyzing execution profile.<br/>The ridge orientation estimation method introduces an original computing method, which is adapted to the specific precision needs and saves the use of high computational cost operations. A numerical value, indicative of the estimation reliability, is associated to the computed orientation. This value will be used to simplify the execution of a fingerprint segmentation step, previous to the feature extraction. Usually this step has been carried out as an independent part of the process with the consequent increase in the total computational cost.<br/>With the presented set of functions and algorithms, and their hardware counterparts (hardware software co-design), it is developed an electronic device with little size, low cost, and high quality results. As a result, the thesis brings new application fields for the personal identification based in fingerprint biometry.
Repositori URV