Our study mainly focuses on real time fingerprint recognition system. The main contributions of this paper are: Firstly, we promoted a combined gray-scale and frequency feature based fingerprint image quality assessment algorithm. In this algorithm, we acquired the image quality proto-type by cluster the contrast, variance of gray-scale, ridge gradient coherence, and dominate component in frequency space. Then we do the fingerprint image quality assessment through the KNN with systems. Experiment results show that the algorithms do correct assessment no less than 94% in Pattek fingerprint Database. Secondly, we promoted a multi-resolution direction field based singular point detection algorithm. The algorithm can precisely localize the singular point, meanwhile need not any post-processing. We first detect singular point on low-resolution direction field, then in the detected area use the high-resolution direction field to precisely localize the singular point. The algorithm can localize the singular point successfully above 98% on Pattek fingerprint Database. Thirdly, we promoted a novel direct gray-scale fingerprint feature extraction algorithm. In this algorithm, we presented the ridge line tracking, initial tracking points select scheme, ridge line tracking ending conditions and minutiae classification Experiment results show that the promoted algorithm is a high-precision and efficient feature extract method. The feature extraction precision is 88% in Pattek Fingerprint Database. Finally, we promoted a core-based structure minutiae matching algorithm. In this algorithm, we first construct some local structure for the minutiae around the core point, then we acquired the corresponding point pair through match those structures. Using those corresponding point pair, we do the global minutiae matching. Finally, we verify the match result through the global match distance and variance. Experiment result shows that the False Match Rate less than 0.1% and the False Not Match Rate less than 1% on Pattek fingerprint database. In order to dual with the deformity fingerprint which does not includes core points, meanwhile keep the performance and efficient of our first matching algorithm, we promoted a globaldistance based fingerprint minutiae match algorithm. This algorithm use the INN to find the corresponding point pair, then do global matching, finally make the decision based on global match error. The average Equal Error Rate is 6% on the FVC2002 fingerprint database. Parts of the algorithms mentioned in this paper have been used in the DSP based Fingerprint Recognition Module developed by Beijing Pattek Ltd.
修改评论