Neural Network-based Fingerprint Identification System

In recent years, unreliable access authorization caused much business and individual loss in modern society. Biometric, instead of password, is often provided as a promising solution to this safety problem. Fingerprint recognition, one of biometric techniques, is popular and reliable for automatic personal identification. This technology has been studied for many years and many methods have been proposed to deal with fingerprint images. And fingerprint recognition has been widely used to identify a person in many areas, such as forensics, genetics, civil, commercial and government.The main steps in a fingerprint recognition system usually include: fingerprint acquisition, image preprocessing, feature extraction, classification and matching. In this thesis, a designed on-line Automated Fingerprint Identification System (AFIS), which is based on global features and minutiae using neural networks, is proposed. The aims of this system are:1. On the basis of some existing fingerprint algorithms, optimize them in order to get better results, especially used on low-quality fingerprint images;2. Apply neural network models to process fingerprint images.3. Design fingerprint system software including fingerprint acquisition, processing, recognition, identification, algorithm testing, system evaluation, etc.This thesis works on some important fingerprint topics, such as fingerprint preprocessing, fingerprint feature extraction, fingerprint classification, fingerprint matching and system design.1. Fingerprint image preprocessing. This thesis proposes a normalization operation. And pulse coupled neural networks (PCNN) is discussed to thin a fingerprint image; then the post-processing of thinning is proposed to deal with unnecessary breaks and links.2. Fingerprint feature extraction. PCNN is used to extract main ridge, then both the main ridge projection and angle nearness are used to estimate fingerprint orientation. Poincare index is used in malposed fingerprint blocks of different size to compute the singulars. Moreover, both Template method and Crossing Number method are used to extract minutiae. How to remove false minutiae and the minutia reliability problem are discussed.3. Fingerprint classification. Singular-based fingerprint classification is proposed in this thesis. And several dynamic masks have been introduced to classify fingerprint images into one of the five classes using the fingerprint orientation.4. Fingerprint matching. Because of translation, rotation and scaling, a matching algorithm is divided into two stages: alignment stage and matching stage. Firstly, the minutia associated with ridges is used to align the template fingerprint image as well as the input fingerprint image, and then, the local comprehensive minutia relation is used for alignment, too. At last, a matching method based on minutia string is proposed.5. Fingerprint recognition system design.