Pulse Coupled Neural Networks And Their Applications In Fingerprint System

Biometrics is a class of technologies that uses person's biologic features, such as face and voice, to verify a person, herein fingerprint is one of these widely adopted biologic features. Fingerprint recognition system often includes some important processing steps, such as fingerprint enhancement, orientation field computation, feature extraction, minutiae matching and pattern classification, etc. In recent years, the technologies and applications related to fingerprint recognition have obtained tremendous progress; nevertheless, there still exist many unsolved problems. These problems could make the application fields of fingerprint recognition system to face many restrictions. Many organizations are still addressing the research on novel technologies and applications for fingerprint, and they have achieved much valuable harvest, and the harvest could promote the application position of fingerprint in the person identification by biometrics.In the beginning, this paper makes a brief review on the existing pulse coupled neural networks (PCNNs), and then, proposes several modified PCNN models, such as the weighted linking PCNN (WL-PCNN), the improved PCNN (I-PCNN) and template-based PCNN (TB-PCNN). On the basis of such models as these, it develops some new image processing algorithms, such as the mixed noise removal. Moreover, this paper researches on the high-dimensional models, as a result, it completes a universal model that is compatible with the traditional two-dimensional ones, then it exhibits some simulations of the pulse characteristics.In addition to the above, this paper proposes some new processing algorithms for the important modules in fingerprint recognition system, such as fingerprint ridge enhancement, fingerprint orientation field computation, binary fingerprint image thinning and fingerprint pattern classification. On the basis of such algorithms as these, this paper designs an integrative fingerprint system, which consists of the client station software (named as FS-Client) and the central service software (named as FS-Server). In this system, the algorithm libraries can be extended, and the fingerprint processing flow can be customized flexibly. The system can be taken as a testing platform of algorithms, as well as a special application system by customization, for example, a door-controlling system. As a whole, this paper mainly includes these parts: the first chapter gives out the research background and significance of fingerprint technologies, and discusses current research situation at home and overseas, the fingerprint algorithms and their evaluation rules, as well as the pulse coupled neural network that frequently appears in the following chapters. The second chapter makes some modifications for the existing PCNNs, and then develops several image processing algorithms using these modified models. The third chapter proposes a fingerprint enhancement algorithm using the modified model, and exhibits some experimental results of such an approach. The fourth develops a novel method for fingerprint orientation field computation, which determines the primary ridge by a simplified PCNN, then computes the local directions by projective distance variances, and makes a correction for the initial orientation field by a low-pass filtering. The fifth chapter proposes a coarse-to-fine thinning algorithm for binary fingerprint image. This method can restrict the thinning direction by orientation field, so it seldom generates pixel spurs. The sixth chapter proposes two fingerprint classification approaches. One is based on PCNN and learning vector quantization (LVQ) networks, and the other is based on a tree-like hierarchical SVM (support vector machine) classifier. The seventh one designs an integrative fingerprint system platform. In the last chapter, we summarize the work of this paper and give out some hot topics for the future research on biometric and fingerprint system.