Research Of Automatic Fingerprint Identification System Based On The Large Capacity Fingerprint Database

Fingerprint recognition technology is one of the focuses in the personal identification field based on the biological features. Although the research of its theory and application has achieved some results, automatic fingerprint identification system based on the large capacity fingerprint database is still insufficient on recognition accuracy rate, speed, robustness and so on. This paper has researched several key algorithms involved in the system comprehensively. In addition, a system development platform has been designed. The main research works are summarized as follows:1. In the aspect of fingerprint image preprocessing, firstly, extracting the grayscale average, direction signal variance and spectrum energy ratio of sub-block image as the feature vector, a linear classifier used for separating the foreground and background regions of the fingerprint image has been designed, which can reduce the segmentation error rate and improve the running speed. Secondly, a fingerprint image quality evaluation method based on independent characteristics has been researched, which using the effective area of the image, the grayscale contrast and spectrum energy ratio of the effective area to evaluate the fingerprint image quality in turn. The low quality image will be refused, and the user will be prompted to re-capture the fingerprint image.2. In the aspect of fingerprint feature extraction, firstly, two-steps positioning method to extracting singular points has been studied, which is on the basis of the Poincare index value of the fingerprint image orientation field and complex filter. The position and direction of singular points can be calculated precisely. This method can extract the information of singular points accurately, and be insensitive to noise. Secondly, a new approach to the minutiae extraction is also put forward, which is on the basis of identity detection. After enhancing the same original fingerprint image respectively by wavelet-based Gabor filter and frequency domain filter, minutiae of two different preprocessed fingerprint skeletons are extracted, then post-processing the union of two sets of minutiae. This approach can eliminate the false minutiae which are inherent in the fingerprint and produced in preprocessing steps, and improve the accuracy of minutiae extraction greatly.3. In the aspect of fingerprint classification and matching, firstly, two-steps fingerprint classification algorithm has been researched. After using a multi-class support vector machine classifier based on grain type to classify fingerprint roughly, fingerprint is classified in details according to the number of ridges between singular points, in order to retrieving and matching rapidly. Secondly, a multilevel fingerprint matching method has been presented. After determining the subclass of the fingerprint being tested in fingerprint database, the registration is carried on according to the local structure of minutiae and direction field, then the final matching result is obtained. This method can improve the matching speed and recognition accuracy in the large capacity fingerprint database effectively.4. Combining the object-oriented idea with database technology, a system development platform has been constructed by using Visual C++6.0and SQL Server. The system integrates a variety of recognition algorithms and establishes a fingerprint database, in order to researching the algorithms of fingerprint recognition systematically.