The Research Of Low Quality Fingerprint Enhancement And Matching Algorithm

The research on fingerprint recognition as a scientific project began in the 19 century, and has matured as an important biometric recognition technology in many application areas. In the meantime, fingerprint recognition technology itself still faces big challenges. On the one hand, in the practical situation, there're about 10% low quality fingerprints which heavily affect the performance of the current automatic fingerprint recognition system. On the other hand, the growing requirements and raising standards in the future urge the fingerprint recognition system towards high performance of reliability and stability.Therefore, with respect of the long term development and application of fingerprint recognition technology, the research on low quality fingerprint recognition remains an important issue.The core procedures of automatic fingerprint recognition system include three parts which are fingerprint preprocessing, minutia extraction and minutia matching. This paper focuses on fingerprint enhancement which is a crucial algorithm of preprocessing, and minutia matching. The main work is detailed as below:1. The theories of several representative fingerprint enhancement algorithms has been deeply studied on, including the directional Fourier filter, the Gabor filter, enhancement algorithm for the singular area based on Poincare index, etc.The limitations of L. Hong's enhancement algorithm based on the Gabor filter has been deeply analyzed, through which a new algorithm based on the theory of short time Fourier transform and probability has been proposed. The new algorithm is able to calculate the crucial parameters, orientation and frequency, simultaneously for the Gabor filter. The algorithm has been simulated and applied in the FVC2004 low quality fingerprint database. The experimental result manifested a high accuracy rate for fingerprint recognition and indicated an obvious reduction of equal error rate from 12.5% to 11.5%.2. The theories of several representative minutia-oriented fingerprint matching algorithms has been deeply studied on, including the algorithms based on triangles, the bozorth algorithm, the k-plet algorithm, etc.The k-plet algorithm based on the graph theory and breadth first search algorithm has been simulated and applied to FVC2004 low quality fingerprint database. The experimental results manifested a 12.7% equal error rate, which indicated an obvious advantage over the bozorth algorithm whose equal error rate was 13.4%.