Study On Key Problems For Automatic Fingerprint Identification System

Biometrics, which refers to identifying an individual based on his or her physiological or behavioral characteristics, is inherently more capable and more reliable in differentiating between an authorized person and a fraudulent imposter than traditional methods such as token-based method (passport or driver license) and knowledge-based method (password or personal identification number, PIN), and has been extensively researched and used. The Automatic Fingerprint Identification System (AFIS) is in more and more people's good graces and has became one of the most important biometrics since for its small size, low cost, easy manipulability and high reliability. In this paper, we focus our research on the fingerprint image segmentation, fingerprint's orientation field computation, fingerprint's singular point detection, fingerprint image thinning and fingerprint matching, and receive the following achievements:1, As for the fingerprint image segmentation, the research work mainly includes two parts: 1) proposes a new segmentation feature called Effective Point Cluster Degree (EPCD)for fingerprint images; 2) proposes an effective method for the segmentation of fingerprint images based on the new feature and the proposed CluD feature, in this new method, we first segment fingerprint images with EPCD, and then process the first segmentation results with iteration-based method , do the second segmentation with CLuD based on the processed results, and finally, morphology has been applied as post processing to reduce misclassification. All experiment results show that: compared with other commonly used features, the EPCD holds the characteristics of good discriminability, robustness, and the segmented foreground and background are more concentrated; at the same time, the segmented method based on EPCD and CluD possesses of high accuracy and adaptability.2, As for the computation of fingerprint's orientation field, our works mainly includes two parts: 1) analyzes and researches three key points of the gradient-based fingerprint's orientation field computing method, which are point gradient vector normalization, block size choosing and promoting the ability of robustness against noises, and proposes the solutions for each key point; 2) systematically proposes two improved gradient-based fingerprint orientation field computing methods based on the above solutions, which are method based on multi-scale windows combination and method based on composite window. Experiments prove the effectiveness of our methods.3, As for the detection of fingerprint's singular point, we introduce a new feature called Orientation Abundance Degree (OAD) based on fingerprint's orientation segmentation, and propose a new method for rapid detection of fingerprint's SP by using OAD. In this method, fingerprint's orientation is partitioned into a series of non-overlapping homogeneous areas firstly; then SPs are rapidly localized through edges detection of homogeneous areas and end-points extraction of edges; finally, the types of SPs are distinguished according to the characteristic of OAD. Compared with the Poincare Index method (PI), experiments on FVC2002 show that: for the performance of accuracy, the False Detecting Ratio (FDR) of our method is much lower than PI's FDR, and the Missed Detecting Ratio (MDR) is a little higher than PI's MDR; for the performance of practicality, our method owns obvious superiority, the speed of our method is 17.4 times faster than PI.4, As for the fingerprint image thinning, we propose a rapid fingerprint image thinning method base on the optimized templates to solve the contradiction between thinning quality and thinning speed for the already proposed One Pass Thinning Algorithm (OPTA). The new method mainly includes two works: proposes the combination templates based on the elimination templates and preservation templates to speed up the templates matching processing effectively; optimizes the combination templates to solve the contradiction mentioned above. All experiments show that: our method guarantees the thinning quality while the thinning speed is 3～6 faster than the proposed methods.5, As for the fingerprint matching, our work mainly includes two parts: 1) discovers that the Orientation_based Minutia Descriptor (OMD) is rotation-interrelated and solves the problem by improving the OMD similarity degree computing method; 2) proposes a fingerprint matching scheme based on multi-stage validation mode, and implements a case, the main procedure is: performing the local matching based on OMD and obtaining the coarse level correspondence set; constructing the higher local topological structure to validate the coarse level correspondences hierarchically, and obtaining the refined correspondence set; for the correspondences which belong to the coarse level correspondence set but not belong to the refined correspondence set, constructing the higher local topological structure based on the refined correspondence set to validate them further, and obtaining the final correspondence set; making the matching decision based on the correspondence count of the final correspondence set. Experiments prove the effectiveness of our scheme.