9 » IC Electronic information » Category: A

IC Electronic information

A Hybrid Fingerprint Matching Method

In Electronic Infomation Category: A | on April 28,2011

Abstract: In order to overcome the traditional minutiae-based matching is insufficient, based on point pattern matching algorithm and CY7C63001A-PC datasheet and improved recognition 2DPCA hybrid matching algorithm is improved. Improved algorithm for point pattern matching algorithm by adding an improved algorithm for the initial matching scores 2DPCA weight, improve the accuracy of point pattern matching algorithm; and CY7C63001A-PC price and the use of point pattern matching algorithm to match the results of the algorithm 2DPCA second match, but also 2DPCA algorithm improves the matching accuracy.

Fingerprint identification technology is an important biometric technology is widely used. Fingerprint recognition generally include fingerprint image acquisition, fingerprint image enhancement, feature extraction and CY7C63001A-PC suppliers and matching parts, which feature match in the whole system occupies an important position. Fingerprint matching algorithm in the system is divided into detail and based on two models based on global information.

Present, most fingerprint identification systems are based on minutiae matching method that extract the fingerprint image after thinning the endpoint and bifurcation point information, the use of some matching algorithm. Although these algorithms have achieved good recognition effect, but shifted, deformation, broken lines and other low quality fingerprint image results are poor, and the characteristics of such methods in the extraction of the fingerprint image before doing a series of pre- treatment, time consuming.

This paper, a model based on point algorithm and improved 2DPCA hybrid matching algorithm that can take full advantage of the fingerprint ridge and valley in the ridge line of the global information, to make up for lack of point pattern algorithm.

1 point pattern matching algorithm based on

This paper, a central point in the polar coordinates of the fingerprint matching algorithm based on the specific steps of the algorithm is as follows.

(1) construct a collection of fingerprint image features in the fingerprint image pre-processing stage to calculate the feature points and feature points, including the coordinates of feature points FeatureX and FeatureY, the direction of the feature points and feature points of the type DirectiON Type (including endpoints and branching points). Through this information, set up the system database already exists in the fingerprint image is P, the number of feature points m, online entry of the fingerprint image as Q, the number of feature points, n, then their corresponding set two points as:


(2) access to the center of CorePoint_X, CorePoint_Y and the direction of the center CorePoint_Dir.

(3) The center of the image in their respective poles, according to the formula all the feature points are converted to polar coordinates:

Converted by the above formula, the fingerprint image in any one feature point can be expressed as a four dimensional vector (Radius, , Dir, Type). Which, Radius said that the feature points in the polar coordinates of the polar radius, the polar angle, said, Dir indicates that the feature points in the direction of polar coordinates; Type indicates the type of feature points.

(4), respectively, the fingerprint template feature points P and Q enter the polar angle in accordance with the direction of increasing order, the formation of two new feature point set:

(5) Set the value of matching error. In order to overcome the fingerprint rotation occurs, the deformation caused by the nonlinear deformation of the error, this paper introduces the concept of variable boundary box, as shown in Figure 1, where, Rw between two feature points for the polar radius allowable error range, w for the polar angle between the allowable error range.

Figure 1 Schematic diagram of the variable boundary box

Far from the center of the feature points of a possible significant reduction in displacement or deformation, and from the center of the feature points tend to occur near the small displacement or deformation, in order to reduce false positives, will be set Rw and w two dynamic values, the specific value of the polar radius by a different decision. Also features

match point to set the direction of the direction of error range Dw, thanks to a discrete eight directions, so the range is Dw = {Dir-1, Dir, Dir +1}, where Dir = 1 Shi when , Dir-1 = 8; when Dir = 8 Shi , Dir +1 = 1.

(6) sorted the input point set Q of the point set of feature points and the template feature points P in each match. When the input image and the template image in more than 13 pairs of feature points satisfy the condition, then that these two fingerprints from the same finger, matching success; the other hand, failed.

2 2DPCA based on an improved fingerprint identification

2DPCA algorithm is an analysis of the object in image feature extraction algorithm, so the image covariance matrix is ??constructed, you can directly use the image matrix. 2DPCA algorithm to image the global information was treated in the realization of dimensionality reduction and feature extraction process, given the same image matrix of each pixel position, if the direct use of image processing algorithms 2DPCA, will inevitably lose part of the between-class discriminant training samples contain information.

Less than the above, this type of design information based on the sample improvements 2DPCA algorithm based on the sample type of information, diversity, within the class using the sample covariance matrix as the feature vectors of the generator matrix, the use of clustering value vector and between-class covariance matrix to extract the characteristics of the training samples. 2DPCA

2.1 Improved algorithm Suppose the training samples

m n image matrix, the total number of P, the number of categories of training samples, L, l-class set number of training samples Pl, then the meeting:

A piece for the first class of training samples l X , the projection space U, the X projection to U will produce a projection matrix Y = XU. With the projection Y total dispersion as a criterion function J (U) to measure the projection space U merits, the criterion function to meet:

Which, SU is the projection matrix Y = XU covariance matrix, tr (SU) for the SU of the track. Number Pl for the first class of sample images l xi (i = 1,2, ..., Pl), the average class can get the sample images to meet:

Using equation (7) of the sample class to mean all of the images:

By their covariance matrix satisfy:

Samples in the category in the obtained covariance matrix G , the calculation of the eigenvalue matrix and eigenvector matrix. The characteristic value of such samples is the eigenvalue diagonal matrix elements, while the corresponding eigenvectors obtained. For each type of sample, whichever is the characteristic value of k before the corresponding eigenvectors as the projection space U :

UiTUj = 0; i j; i, j = 1,2, ..., k so that l can come to the first class of sample images Xi (i = 1,2, ..., Pl) in the space U in the projector to meet:

The Yi is the original image in this category xi post-dimensionality feature vector, as this type of image projection vector matrix, the sample used to identify the image type. Similarly, the L class of training samples increase total P were trained by the sample type, can be a projection vector matrix L.

2.2 Improved algorithm for fingerprint matching 2DPCA

Fingerprint classification, will be the effective area of ??training samples extracted by four new sample set. Then deal with each type of training samples were obtained after the projection feature vector.

Line input for the test samples, the same to get it in the space U of the projection vector. Suppose T is a test sample image to be identified, after a judge determined after the sample type l T is the first class, that is, T Pl, using equation (10) go to the mean:

Be projected into the feature space, by (11) are the projection of the input sample vector:

With the projection vector Yt

classification of training samples increases the projection Pl vector Yi to match the distance, in accordance with (12) to calculate the Euclidean distance:

Finally nearest neighbor rule, when the sample such as a T with a similar piece of training samples Plj (Plj Pl) has the smallest Euclidean distance and the distance to meet a certain threshold when the input samples to determine T and training samples for the same image, that is to complete the identification.

3 mixed model-based fingerprint recognition algorithm

Based fingerprint recognition algorithm for mixed-mode flow chart shown in Figure 2.

hybrid pattern matching algorithm in Figure 2 Flow Chart

Set to N rate were collected fingerprint image, the sample is divided into K classes, where the first k (k [1, K]) class contains M images, the concrete realization steps are as follows:

(1) input fingerprint image acquisition and quality * estimates; (2) samples of the input fingerprint image type, set the input belongs to the first k classes; (3) 2DPCA the input fingerprint image pre-processing; ( 4) to extract the input image feature vector set 2DPCA; (5) using 2DPCA fingerprint image matching algorithm in the k-class database for the early match, if not meet the matching requirements, the system eventually fails to match; met, through the appropriate threshold set by m (m "M) piece matching candidate fingerprint and their score weighting, and also according to the index to get their point model feature points; (6) points on the input mode of fingerprint image preprocessing; (7) on the pre- after the input fingerprint image features focus point mode, point pattern matching algorithm using the second match, and add the corresponding match score 2DPCA weight. If you meet the matching requirements, the system will eventually match the success; if not satisfied, then fail.

4 Experimental results and analysis This article is

CPU 2.00 GHz, 1.99 GHz, 2.00 GB of memory for the PC and Matlab R2007B, Visual STudio 2007 development environment, the choice of FVC2002DB2_A of 880 fingerprint image matching algorithm experiment. The fingerprint database of 110 fingerprints were collected, 8 were collected for each finger are eight fingerprints. Cross-match test using the manner in which each finger from the eight selected 6 as the template fingerprint, two input fingerprints, a total of 220 times to match the experimental results are shown in Table 1.

Can be seen from the table, using this algorithm matches the fingerprint recognition rate of 93.57%, and the point pattern matching algorithm, the recognition rate increased. 2DPCA algorithm improved access features in the space area of ??Viti Levu You Yu Xiang fingerprint image to a larger degree of displacement, and some of the more serious the phenomenon of adhesion, so the final algorithm is obtained based on the principle of nearest neighbor matching image appears an error, but to observe the Euclidean distance value of the order, the right general in the forefront of the fingerprint image, which is the basis for hybrid matching algorithms. Mixed matching, the recognition rate slightly improved. This article will point pattern matching algorithm combined with 2DPCA in point pattern matching algorithm is added to the initial matching algorithm scores 2DPCA weight, improve the accuracy of the point pattern; and category information based on the method of the sample, greatly reducing the point pattern matching the original data points in between the search and by the number of matches, so a little bit better than the original pattern and high efficiency.

three models in Table 1 fingerprint matching algorithm results

This point based on the details of the fingerprint matching algorithm and improved global information-based matching algorithm was analyzed 2DPCA; then the algorithm of three models were compared, summed up the advantages and disadvantages; Finally, the two modes of algorithms designed by combining a hybrid fingerprint recognition algorithm. The algorithm has the advantages of both models can reduce the matching and reduce the number of matches, and to some extent, improve the recognition rate and reduce false positives and rejection rates.

CY7C63001A-PC datasheetCY7C63001A-PC suppliersCY7C63001A-PC Price

All right © 2010-2016 Certificate