| The identification and consistency of fingerprints together with the maturity and reliability of identification technology make it been used in more and more fields such as social security, office security, information security, finance security and personal security. The fingerprint identification technology has developed very quickly and has a bright market prospect. The huge amount of data in large fingerprint databases (several million fingerprints) seriously compromises the efficiency of the fingerprint identification task in Automated fingerprint Identification Systems (AFIS) for both forensic and civil applications. Adopting a classification approach is a common strategy to reduce the number of comparisons during fingerprint retrieval and, consequently, to improve the response time of the identification process.An automated fingerprint identification algorithm should quickly classify the fingerprints at a satisfied accuracy. But in practice, due to the variety and complexity of the fingerprints configuration and the poor quality, fingerprint classification faces a great challenge. Though a large number of papers have been published on this topic during the last 30 years, automatic fingerprint classification performance is currently below operating requirements. Fingerprint classification problem has always been a difficulty in the AFIS.The algorithm of this paper was proposed to speed up the on-line identification process for the civil applications. It may not be fit for other uses, such as forensic applications.Innovation of the paperDesigned for the civil fingerprint classification. The strategy of the technique of the paper is to search the input fingerprint in the hypothesized class only. Obviously this assumes no classification errors, which is quite unlikely for state-of-the-art automatic classifiers. So the algorithm of this paper assigned each fingerprint from one to three classes. A fingerprint to be identified is then to be compared only with the fingerprints in a single bin of the database based on its classes. For every classifier suffers from the partial fingerprints or the poor quality fingerprints. As for the most total fingerprints with good quality, we can identify its class through some approach. In other retrieval strategy, if a correspondence is not found within the hypothesized class, the search continues in another class, and so on. Such a lot of time is wasted when compared two total fingerprints with sure different classes. The approach of the paper is rule-based. A fingerprint is classified according to the core and the lines tracing from near the core. The lines are traced on the 16×16 block dir image and consist of the blocks. The coarse classification is completed on the bases of ending condition of the lines and the curvature of the lines. This not only allows reducing noise and consequently improving classification accuracy but also speeds up the process of the classification. So it well suits the civil applications. The retrieval strategy assumes high accuracy of the classification. So we assign the twin loop into the correspondence loop class. For the tented arch we assigned it to the left loop and the right loop. The performance of this method shows that it can avoid the usual errors of other algorithms.The sub-classification is completed according to the distances between the two cores, between the core and delta and between the core direction and the direction of line from the core to the delta. If each is larger than the given criterion, the two fingerprints with the same class are not going to be matched. The sub-classification is also a remedy for the coarse classification. As for the tented arch, though we trace three lines near the core, sometimes three lines all turn to the left or the other side and the error occurs. So when it comes to the sub-classification, if the fingerprint is classed into the tented arch according to the core and delta, we supplement the other class. The multi-verification of the singular...
|
PDF Full Text Request