In today's complex, geographically mobile and increasingly electronically inter-connected information society, accurate personal identification is becoming more and more important and difficult. The use of fingerprint as a biometric is both the oldest mode of automatic personal identification and the most prevalent in use today. This thesis focuses on algorithms related to Automatic Fingerprint Identification Systems (AFIS), and the main contributions are as follows: 1. A comprehensive survey is presented on the state of the art of the core algorithms related to AFIS as well as some system issues (such as the architecture and technical evaluation of biometrics system). 2. The ultimate objective of fingerprint image analysis is to achieve the most reliable features and can be divided into two consecutive parts: pre-processing and feature extraction. In pre-processing, we proposed a projection analysis method to establish a reliable estimation of ridge width map. Based on this information, we compared the performance of three typical fingerprint filters and meaningful results are obtained. As to feature extraction, we developed an effective implementation of fingerprint post-processing, which meets the accuracy and response-time requirements of real-time systems. Finally, we proposed a singularity detection algorithm which integrate, s local orientation variance and Poincaré Index. 3. A novel error propagation based fingerprint matching algorithm is proposed, which is capable of adaptively tracking the nonlinear deformation commonly observed in fingerprint images. The main idea of this approach is to estimate the errors of the unmatched minutiae according to those of the matched minutia pairs. To prevent the matching procedure from being misguided by mismatched minutia pairs, a flexible diffusion scheme is developed. The EERs of 2.05% and 1.5% are obtained on NIST-24 and NLPR databases respectively. 4. A prototype multi-biometric identity verification system, which integrates both voice print and fingerprint information, is developed. The outputs of two subsystems are regarded as a two-dimensional feature vector and linear discriminant function based on Fisher criteria is used in the integrated classifier. Experimental results demonstrate that our decision fusion scheme performs well.
修改评论