Fingerprint recognition has become the most widely applied and inexpensive biometric identification technology. In order to solve the problems in the application of fingerprint recognition, we, in this thesis, focus our study on several key models of automatic fingerprint recognition system, such as fingerprint image quality evaluation, fingerprint registration, fingerprint classification, fingerprint feature extraction and matching. The main contributions of this thesis include: 1. According to the requirement in real application of fingerprint identification system, an automatic quality evaluation algorithm of fingerprint images is proposed. The method integrally evaluate the quality of captured fingerprint images in several respects, such as fingerprint foreground area, pressing position of finger, gray-level variation of fingerprint image, orientation consistency and clarity degree of ridge-valley structure. The algorithm can accurately prompt the user for different kinds of bad quality fingerprint images in the process of fingerprint capture, and can also mark the region of bad quality to ensure reliability of fingerprint feature matching 2. A two-stage singular points detection method, which includes coarse locating stage and precise locating stage, is proposed. And then a reference point based fingerprint image registration algorithm is put forward to effectively align the fingerprint images, which can improve the accuracy of fingerprint matching. 3. An automatic fingerprint classification method based on gray-level statistical features and singular point information is proposed. The accuracy of 93.23% with no rejection and 99.32% with 9.8% rejection respectively was achieved on NIST-4 database. Experimental results show that the proposed method, which can greatly reduce the size of the search space and accordingly increase the matching speed of a large-scale fingerprint database, is feasible and reliable for fingerprint classification. 4. An in-depth study was conducted on fingerprint recognition based on image features, including gray-level statistical feature, texture feature based on Gray Level Co-Occurrence Matrix，and histogram feature based on Local Binary Pattern Operator. Experimental results evidence that the 3 proposed methods have higher accuracy and much faster speed than the most famous non-minutiae based method “FingerCode” based on Gabor filter. 5. A fingerprint recognition algorithm is put forward based on the fusion of different fingerprint recognition methods, including process of fingerprint classification, classifier selection and classifier fusion. Four image-based methods and one minutiae-based method are combined for fingerprint recognition. Experimental results indicate that the performance of fusion-based method is better than any method used alone in fusion and can be applied to automatic fingerprint identification system.
修改评论