Research Of A Alignment-free Fingerprint Phase Feature Extraction Algorithm

With fingerprint recognition technology increasingly wide range of applications, and on the capacity fingerprint database requirements increasing, collected fingerprint image showing the diversity of characteristics than conventional systems than for a long time, there is an urgent need for an easy-toratio Yes, high recognition accuracy, small storage space occupied by fingerprint characteristics.This paper presents a frequency domain fingerprint-free calibration phase characteristics, and according to the extracted feature built new automated fingerprint identification system to optimize each processing module of the system. In the pre-processing module, based on the two-dimensional mathematical model of a fingerprint, a fingerprint using the same of different pieces of fingerprint image with the same phase characteristics, time domain and frequency domain binding two kinds of the treatment space, through the filtering to eliminate noise and compensation term in the model, the use of Fourier spiral phase operator and Euler's formula, carry out modulation and demodulation, extract the phase term included in the model features. In the frequency domain feature extraction module, since the acquisition and collection of different finger pressure position, so that the collected fingerprint image will appear translation, rotation and scaling, therefore, of the preprocessed fingerprint image, according to the Fourier transform of the translation translation invariance eliminate the influence of the image; because the fingerprint feature mainly in the lighter areas of the spectrum, with a sample standard deviation to determine the division range, the spectrum of useful and useless region divided region; by a logarithmic polar coordinate transformation, Cartesian coordinate system rotation and scaling of the number is converted to polar coordinates translation, re-use of the Fourier transform of the image rotation and translation invariance eliminate the impact of scaling; use linear interpolation to eliminate the influence of the image after smoothing. By Matlab build Automated Fingerprint Identification System, feature extraction and matching, validation algorithm, to real fingerprints and false fingerprint test samples were experiments to obtain ROC curve in the evaluation of system performance, according to the position of equal error rate point analysis system feasibility.Characteristics of this paper is not affected by the position, direction, so no calibration can be compared directly, reducing system than in the matching part of the time, to ensure recognition accuracy at the same time, effectively improve the recognition speed of the entire system.