Design of multiple enrollment based fingerprint recognition systems

Abstract

Using multiple enrollment can improve recognition performance in fingerprint recognition systems; but there are several technical and operational challenges to implementing multiple enrollment based fingerprint recognition systems. Multiple enrollment based fingerprint recognition systems still have low recognition accuracies, poor matching speeds, and consume a lot of memory making it difficult to implement them in real world scenarios. Also, most of multiple enrollment based fingerprint recognition systems have been designed mainly based on minutiae approaches but not others such as correlation and pattern based approaches hence limiting implementation. The purpose of this research was to provide a novel multiple enrollment fingerprint recognition approach that further improves recognition accuracy, the matching speed and reduce memory consumption in multiple enrollment based fingerprint recognition systems as well as allow for implementation using non-minutiae methods. In this thesis, a literature survey of the state of the art in multiple enrollment for fingerprint recognition was first performed. A list of laboratories working on multiple enrollment for fingerprint recognition was also generated. This literature survey serves as a quick overview of the state of the art in multiple enrollment for fingerprint recognition for the past two decades. This thesis evaluates the effectiveness of using multiple enrollment in fingerprint recognition systems. A Spectral minutiae based multiple enrollment algorithm was designed and used together with existing fingerprint recognition techniques to carry out the evaluation. The experimentation results and evaluations show that multiple enrollment as whole outperforms single enrollment. Multiple enrollment in experiment one improved the recognition performance by 83.33% from EER of 0.75% to EER of 0.13% with FVC2000-DB2 fingerprint database, and by 75.55% from EER of 1.14% to EER of 0.28% with the SAS-DB2 fingerprint database. On the other hand, the multiple enrollment in experiment two improved the recognition performance by 71.51% from EER of 6.14% to EER of 1.75% with the FVC2000-DB2 fingerprint database and improved recognition performance by 53.61% from EER of 14.97% to EER of 6.94% with SAS-DB2 fingerprint database. A comparison with single enrollment and other multiple enrollment results in literature shows that our algorithms were superior by over 38.1% in terms of recognition performance. This research developed a novel approach that performs prior selection of good fingerprint image samples of an individual for matching and further improves recognition performance, reduces the matching xxi speed as well as memory consumption. A spectral minutiae based matching method and two fingerprint databases (FVC2000-DB2 and FVC2006-DB2) were used. A comparison of our results with the existing ones presented in literature showed that they are more superior by over 29.6% with algorithm two (Alg2) and by 100% with algorithm three (Alg3). This makes it possible to design better multiple enrollment based fingerprint recognition systems with a high recognition accuracy, high matching speed and low memory consumption using our approach. This research also experimented with the Gabor filter-based approach; the first of the kind, to implement a verification multiple enrollment based fingerprint recognition system. The Gabor filter-based multiple enrollment fingerprint recognition method was compared with a spectral minutiae-based method using two fingerprint databases; FVC 2000-DB2-A and FVC 2006-DB2-A. Although the minutiae-based method outperformed the Gabor filter-based method, the results attained from the later were promising and were a good basis to further discussions and improvements for implementing Gabor filter-based techniques in designing multiple enrollment based fingerprint systems. This research embarked on a Combined Feature Level and Score Level Fusion Gabor filter-based approach; an advancement of the previous Gabor filter based method. The Combined Feature Level and Score Level Fusion Gabor filter-based multiple enrollment fingerprint recognition method was compared with a spectral minutiae-based method using the same (two) fingerprint databases as in the previous experimentation above. The results indicate that there is a significant percentage increase brought about by the combined feature level and Score Level fusion Gabor filter-based matching approach in comparison to the famous minutiae-based matching approach. The percentage increases in the FVC 2000-DB2-A fingerprint database were 86.45%, 98.01% and 87.82%, while those in the FVC 2006-DB2-A fingerprint database were 79.71%, 97.07% and 85.88% respectively for recognition performance improvement, matching speed improvement and memory consumption reduction. The results attained from the approach above were outstanding and are therefore a proposed possibility for future deployment in real world multiple enrollment fingerprint recognition applications that require better recognition performance, better matching speed and a reduced memory consumption.