Using Neural Networks In Pattern Classification

Abstract

Biometric identification systems have been developed to achieve automatic identification of a person based on his physiological or behavioral characteristics. Bi!Jmetric systems are critical in a wide range of applications such as banking systems. E-commercc. smart cards. and access control to secure systems. Automatic . lit1gcrprinl identification is one of the most reliable biometric systems, which is used for identifying persons. • In this thesis. our objective is to design a fingerprint identification system. 11hich is ce pe hlc.of identifying a lingerprint with high level of accuracy. Therefore. this system e<ln be applied to a wide range of forensic applications.

1 lp this research. we proposed a complete fingerprin identification system, The sy tem is composed of the !<JIIowing stages: (i) Fingerprint image preprocessing, in .; which we proposed a new algorithm to detect the center point of the fingerprint and construct a normalized image for the sectors surrounding the center point. The proposed center point detection algorithm was tested in comparison with previously published algorithms and proven to be more accurate and faster. (ii) Feature extraction, in which we used a bank of Gabor filters to capture both the local and global details of the fingerprint and convert them into a relatively short and fixed length code for the purpose of matching. (iii) Identification using neural networks, in which a back-propagation neural network was used to identify an input fingerprint image fi•01n the system database. We performed several tests using different value Jor the parameters of the neural network in order to reach to the best results. The proposed system has been evaluated on a system database that contains live-scan .fingerprints. The system was able to identify an input fingerprint with an accuracy of %.67%. On the other hand it did not misidentified any input fingerprint. These results were compared with two other identification systems and the proposed system proven to give better results in both the False Acceptance Rate (FAR) and the F lse Rejection Rate (FRR). These results show that the proposed system can be accepted as an accurate identification system.