CN107038432B - Fingerprint image direction field extraction method based on frequency information - Google Patents

Abstract

The invention discloses a fingerprint image direction field extraction method based on frequency information, which is used for solving the technical problem of poor accuracy when the existing fingerprint image direction field extraction method is used for processing fingerprint images with poor quality. Firstly, partitioning a fingerprint image, and respectively performing direct current component removal processing on each image block; performing Fourier transform on the image block from which the direct-current component is removed to obtain a spectrogram, obtaining a corresponding energy map through the spectrogram of the image block, and calculating the energy map to obtain the direction and the frequency of the fingerprint image block; and each image block is corrected according to the direction and the frequency of the surrounding image block to obtain the final direction and frequency. The method can effectively avoid the interference of noise in the fingerprint image, and can obtain an accurate direction field when processing the fingerprint image with poor quality.

Description

Fingerprint image direction field extraction method based on frequency information

Technical Field

The invention relates to a fingerprint image direction field extraction method, in particular to a fingerprint image direction field extraction method based on frequency information.

Background

in automatic fingerprint identification systems, fingerprint image enhancement is a very important step. The key to fingerprint image enhancement is the extraction of the fingerprint image orientation field. Common methods for extracting the orientation field of the fingerprint image include a gradient-based method, a filtering-based method and a model-based method.

The document "Fingerprint Image Enhancement: Algorithm and Performance Evaluation, IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, 1998, 20(8): 777-. The method divides an image into a series of non-overlapping pixel blocks with the same size on the basis of carrying out normalization processing on the image. And (4) calculating the gray level change gradient vector of each pixel point in the block by using a gradient operator-sobel operator. And calculating the gradient vector of the block in which the point gradient vector is positioned by using the obtained point gradient vector. And converting the gradient vector into a block direction to further obtain a direction field of the whole fingerprint image. Finally, the low-pass filter is used for smoothing the direction field to correct the wrong direction. The method disclosed by the document has the core of the calculation of the block gradient vector, the calculation process is easy to generate wrong directions due to the interference of noise in the image, and the robustness is low. Especially when the fingerprint image has scar, fold and other structures, the method can extract wrong direction fields.

Disclosure of Invention

in order to overcome the defect that the conventional fingerprint image direction field extraction method is poor in accuracy when processing a fingerprint image with poor quality, the invention provides a fingerprint image direction field extraction method based on frequency information. Firstly, partitioning a fingerprint image, and respectively performing direct current component removal processing on each image block; performing Fourier transform on the image block from which the direct-current component is removed to obtain a spectrogram, obtaining a corresponding energy map through the spectrogram of the image block, and calculating the energy map to obtain the direction and the frequency of the fingerprint image block; and each image block is corrected according to the direction and the frequency of the surrounding image block to obtain the final direction and frequency. The method can effectively avoid the interference of noise in the fingerprint image, and can obtain an accurate direction field when processing the fingerprint image with poor quality.

the technical scheme adopted by the invention for solving the technical problems is as follows: a fingerprint image direction field extraction method based on frequency information is characterized by comprising the following steps:

Step one, inputting a fingerprint image I (x, y) with the size of H multiplied by W, wherein H represents the height of the fingerprint image I (x, y), W represents the width of the fingerprint image I (x, y), and (x, y) represents the coordinates of the fingerprint image I (x, y);

Step two, carrying out blocking processing on the fingerprint image I (x, y), dividing the fingerprint image I (x, y) into image blocks with the size of m multiplied by n, overlapping two adjacent image blocks, wherein the width of an overlapping area is num pixels, and obtaining a fingerprint image block I_ij(x, y) where I denotes a fingerprint image block I_ij(x, y) is located in the ith row of the fingerprint image I (x, y), j represents the fingerprint image block I_ij(x, y) is located in the jth column of the fingerprint image I (x, y);

Step three, calculating fingerprint image block I_ijThe average value avg of all pixel values in (x, y) is constructed into a matrix V (x, y) with the size of m multiplied by n, the value of each point in the matrix V (x, y) is avg, and the matrix V (x, y) is used for processing the fingerprint image block I_ij(x, y) removing the direct current component to obtain an image D (x, y) with the direct current component removed; performing Fourier transform on the image D (x, y) with the direct-current components removed to obtain a spectrogram F (x, y), and squaring the pixel value of each point in the spectrogram F (x, y) to obtain an energy map E (x, y);

Step four, all energy maximum value points are obtained in the energy diagram E (x, y), all the energy maximum value points are arranged in a descending order according to the energy values, and the energy maximum value points after the ordering are placed in a queue Q_iWhere i ═ 1, 2., N denotes the number of energy maxima in the energy map E (x, y);

Step five, calculating a queue Q_iDirection o of the first two points_iAnd frequency f_iWhere i is 1,2, and setting a frequency threshold Th 1;

Step six, if f₁Not more than the frequency threshold Th1, step seven is executed, otherwise, the fingerprint image block I is processed_ijdirection O of (x, y)₁The size of (i, j) is set to o₁Fingerprint image block I_ijFrequency F of (x, y)₁The size of (i, j) is set to f₁Executing the step eight;

Step seven, if f₂Greater than the frequency threshold Th1, the fingerprint image block I_ijDirection O of (x, y)₁The size of (i, j) is set to o₂fingerprint image block I_ijFrequency F of (x, y)₁(i, j) sizeIs set to f₂Otherwise, the fingerprint image block I_ijThe direction and frequency of (x, y) are respectively set as constant a, a is used for representing fingerprint image block I_ijDirection O of (x, y)₁(i, j) and frequency F₁(i, j) cannot pass through the computation queue Q_iDirection o of the first two points_iAnd frequency f_iObtaining;

Step eight, repeating the step three to the step seven, and solving all fingerprint image blocks I_ijdirection O of (x, y)₁(i, j) and frequency F₁(i,j)；

Step nine, fingerprint image block I_ijDirection O of (x, y)₁(i, j) and frequency F₁(I, j) carrying out first correction to obtain a fingerprint image block I_ijDirection O of (x, y)₂(i, j) and frequency F₂(i,j)；

Step ten, fingerprint image block I_ijdirection O of (x, y)₂(i, j) and frequency F₂(I, j) carrying out second correction to obtain a fingerprint image block I_ij(x, y) final direction O₃(i, j) and frequency F₃(i,j)。

the invention has the beneficial effects that: firstly, partitioning a fingerprint image, and respectively performing direct current component removal processing on each image block; performing Fourier transform on the image block from which the direct-current component is removed to obtain a spectrogram, obtaining a corresponding energy map through the spectrogram of the image block, and calculating the energy map to obtain the direction and the frequency of the fingerprint image block; and each image block is corrected according to the direction and the frequency of the surrounding image block to obtain the final direction and frequency. The method can effectively avoid the interference of noise in the fingerprint image, and can obtain an accurate direction field when processing the fingerprint image with poor quality.

the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

Drawings

FIG. 1 is a flow chart of a fingerprint image direction field extraction method based on frequency information according to the invention.

Figure 2 is an unprocessed fingerprint image.

Fig. 3 is a direction field diagram of a fingerprint image after being processed by the method of the present invention.

Detailed Description

The noun explains:

I (x, y): a fingerprint image;

H: height of fingerprint image I (x, y);

W: width of fingerprint image I (x, y);

(x, y): coordinates of the fingerprint image I (x, y);

num: the width of the overlap region;

I_ij(x, y): fingerprint image blocks;

m: fingerprint image block I_ij(x, y) height;

n: fingerprint image block I_ijA width of (x, y);

N_H: the number of blocks of the fingerprint image I (x, y) in the vertical direction;

N_W: the number of blocks of the fingerprint image I (x, y) in the horizontal direction;

avg: fingerprint image block I_ijAverage of all pixel values in (x, y);

v (x, y): a constructed matrix;

D (x, y): removing the image of the direct current component;

F (x, y): a spectrogram;

E (x, y): an energy map;

Q_i: the energy maximum value points are arranged in a descending order according to the energy values;

Δ x: queue Q_iDistance of the abscissa of the midpoint to the center of the energy plot E (x, y);

Δ y: queue Q_idistance of the ordinate of the midpoint to the center of the energy plot E (x, y);

Q_ix: queue Q_ithe x coordinate value of the ith point;

Q_iy: queue Q_iThe y coordinate value of the ith point;

x_c: x-coordinate value of the center point of the energy map E (x, y);

y_c: the y coordinate value of the center point of the energy map E (x, y);

o_i: compute queue Q_iThe direction obtained from the first two points;

f_i: compute queue Q_ifrequencies obtained from the first two points;

Th 1: a frequency threshold;

O₁(i, j): fingerprint image block I_ij(x, y) an initial direction;

F₁(i, j): fingerprint image block I_ijAn initial frequency of (x, y);

a: marking whether a fingerprint image block I is obtained_ijInitial direction O of (x, y)₁(i,j)；

O₂(i, j): fingerprint image block I_ij(x, y) the direction after the first correction;

F₂(i, j): fingerprint image block I_ij(x, y) the frequency after the first correction;

Th 2: a direction threshold;

O₃(i, j): fingerprint image block I_ij(x, y) final direction after second correction;

F₃(i, j): fingerprint image block I_ij(x, y) the final frequency after the second correction.

reference is made to fig. 1-3. The fingerprint image direction field extraction method based on the frequency information comprises the following specific steps:

Step 1, inputting a fingerprint image.

Taking a fingerprint image I (x, y) with the size of H multiplied by W as an input, in this example, a fingerprint image is randomly selected from a fingerprint image database of FVC (fingerprint Verification) 2002, wherein H represents the height of the fingerprint image I (x, y), W represents the width of the fingerprint image I (x, y), and (x, y) represents the coordinates of the fingerprint image.

Step 2, carrying out blocking processing on the fingerprint image I (x, y), dividing the fingerprint image I (x, y) into image blocks with the size of m multiplied by n, overlapping two adjacent image blocks, wherein the width of an overlapping area is num pixels, and obtaining a fingerprint image block I_ij(x, y) where I denotes a fingerprint image block I_ij(x, y) is located on the ith row of the fingerprint image I (x, y), j denotes a fingerStripe image block I_ij(x, y) is located in the jth column of the fingerprint image I (x, y).

(2a) Calculating the number of blocks of the fingerprint image I (x, y):

Wherein N is_HIndicates the number of blocks of the fingerprint image I (x, y) in the vertical direction, N_Wthe number of blocks of the fingerprint image I (x, y) in the horizontal direction is indicated, and in this embodiment, m is 24, n is 24, and num is 12. If N is present_HFor a decimal value, pixels with pixel values of 0 need to be uniformly complemented at two ends of the fingerprint image I (x, y) in the vertical direction, so that N is enabled_HBecomes an integer. If N is present_Wfor a decimal value, it is necessary to uniformly complement pixels with a pixel value of 0 at both ends of the fingerprint image I (x, y) in the horizontal direction so that N is equal to N_WBecomes an integer;

(2b) Dividing a fingerprint image I (x, y) into fingerprint image blocks I of size 24 x 24 pixels_ij(x, y) wherein i ═ 1,2_H，j＝1,2,...,N_WTwo adjacent image blocks overlap each other, and the width of the overlapping area is num pixels, where num is 12 in this embodiment.

Step 3, fingerprint image block I_ij(x, y) removing the direct current component to obtain an image D (x, y) with the direct current component removed; the image D (x, y) from which the dc component is removed is fourier-transformed to obtain a spectrogram F (x, y), and the pixel value of each point in the spectrogram F (x, y) is squared to obtain an energy map E (x, y).

(3a) Computing an image I_ijaverage of all pixel values in (x, y):

wherein avg denotes a fingerprint image block I_ijAverage value of all pixel values in (x, y), m representing fingerprint image block I_ij(x, y), where m is 24 and n represents the fingerprint image block I in this embodiment_ij(x, y), where n is 24 and mn is the multiplication of m and n;

(3b) For fingerprint image block I_ijSubtracting avg from each pixel value in (x, y) to obtain an image D (x, y) with the DC component removed:

D(x,y)＝I_ij(x,y)-V(x,y)，

where V (x, y) is a 24 x 24 matrix, and the value of each point is avg;

(3c) Calculated energy map E (x, y):

E(x,y)＝F(x,y)·F(x,y)，

Where the sign-denotes the multiplication of the values of the corresponding points in the two matrices.

step 4, all energy maximum value points are obtained in the energy diagram E (x, y), all the energy maximum value points are arranged in a descending order according to the energy values, and the energy maximum value points after the ordering are placed in a queue Q_iWhere i 1,2, N denotes the number of energy maxima in the energy diagram E (x, y).

(4a) For each point in the energy map E (x, y) that matches x 2, 3., 23, y 2, 3., 23, it is determined whether the size of its pixel value is larger than the pixel values of 8 points around this point, if so, this point is the energy maximum value point, otherwise, this point is not the energy maximum value point;

(4b) All the energy maximum value points are arranged in descending order according to the energy size, and the energy maximum value points after being arranged in the queue Q_iWherein i ═ 1, 2.., N.

step 5, calculating queue Q_idirection o of the first two points_iand frequency f_iWherein i is 1, 2.

(5a) Compute queue Q_iDistance Deltax from the abscissa of the midpoint to the center of the energy plot E (x, y) and queue Q_iDistance Δ y from the ordinate of the midpoint to the center of the energy plot E (x, y):

Q_ixPresentation queue Q_iX coordinate value, x, of the ith point_cAn x-coordinate value representing the center point of the energy map E (x, y),In this example x_c＝13，Q_iypresentation queue Q_iy coordinate value of the ith point, y_cThe y coordinate value representing the center point of the energy map E (x, y), y in this embodiment_c＝13；

(5b) Compute queue Q_iThe direction o represented by the point in (1)_i：

o_iPresentation queue Q_iThe direction represented by the ith point in (1), o_iThe value range of (a) is changed to [0, pi), and arctan represents an arctangent function;

(5c) Compute queue Q_iFrequency f represented by a point in (1)_i：

f_iPresentation queue Q_iIn the frequency represented by the ith point, s represents the magnitude of the energy map E (x, y), and in this embodiment, s is 24.

Step 6, setting a frequency threshold Th1, in this embodimentIf f is₁Not greater than a given frequency threshold Th1, step (7) is executed, otherwise the fingerprint image block I_ijDirection O of (x, y)₁The size of (i, j) is set to o₁Fingerprint image block I_ijfrequency F of (x, y)₁the size of (i, j) is set to f₁And (8) executing the step.

Step 7, if f₂Greater than a given frequency threshold Th1, the fingerprint image block I_ijdirection O of (x, y)₁the size of (i, j) is set to o₂Fingerprint image block I_ijFrequency F of (x, y)₁The size of (i, j) is set to f₂Otherwise, the fingerprint image block I_ijThe direction and frequency of (x, y) are respectively set as a, a to represent fingerprint image block I_ijDirection O of (x, y)₁(i, j) and frequency F₁(i, j) cannot pass through the computation queue Q_iDirection o of the first two points_iAnd frequency f_iIn this example, a is-1.

And 8, repeating the steps (3) to (7) to obtain all fingerprint image blocks I_ijDirection O of (x, y)₁(i, j) and frequency F₁(i,j)。

Step 9, for fingerprint image block I_ijDirection O of (x, y)₁(i, j) and frequency F₁(I, j) carrying out first correction to obtain a fingerprint image block I_ijdirection O of (x, y)₂(i, j) and frequency F₂(i,j)。

(9a) judging fingerprint image block I_ij(x, y) is a block of an edge of the fingerprint image I (x, y), i.e., whether I-1 or I-N is satisfied_HOr j ═ 1 or j ═ N_WAny one of the four conditions, if not, executing the step (9b), if yes, calculating the direction O₂(i, j) and frequency F₂(i,j)：

(9b) Judging each fingerprint image block I in the fingerprint image I (x, y)_ijDirection O of (x, y)₁If (i, j) is a, in this example a is-1, and if so, the direction O is calculated₂(i, j) and frequency F₂(i,j)：

Wherein i is 2,3_H-1，j＝2,3,...,N_W-1, c and d are not simultaneously 0;

If not, calculate the direction O₂(i, j) and frequency F₂(i,j)：

wherein i is 2,3_H-1，j＝2,3,...,N_W-1。

step 10, fingerprint map is checkedImage block I_ijDirection O of (x, y)₂(i, j) and frequency F₂(I, j) carrying out second correction to obtain a fingerprint image block I_ij(x, y) final direction O₃(i, j) and frequency F₃(i,j)。

(10a) Judging fingerprint image block I_ij(x, y) is a block of an edge of the fingerprint image I (x, y), i.e., whether I-1 or I-N is satisfied_HOr j ═ 1 or j ═ N_WAny one of the four conditions, if not, executing the step (10b), if yes, calculating the direction O₃(i, j) and frequency F₃(i,j)：

(10b) fingerprint image block I_ijDirection O of (x, y)₂(i, j) is compared to the directions of the surrounding 8 image blocks, where i is 2,3_H-1，j＝2,3,...,N_W-1. Judging the image block I_ijDirection O of (x, y)₂(i, j) whether or not the absolute values of the direction differences from the surrounding 8 image blocks are all larger than a given direction threshold Th2, in this embodimentif so, calculate the direction O₃(i, j) and frequency F₃(i,j)：

Wherein i is 2,3_H-1，j＝2,3,...,N_W-1, c and d are not simultaneously 0;

If not, calculate the direction O₃(i, j) and frequency F₃(i,j)：

Wherein i is 2,3_H-1，j＝2,3,...,N_W-1。

The effects of the present invention can be further illustrated by the following simulations:

1 simulation Condition

The simulation was performed in the matlab2014a environment of a PC, which was equipped with a Core I7 processor and had a dominant frequency of 3.4-GHz. The simulated images are from 7_8.GIF in fvc (finger Verification completion) 2002 fingerprint database DB1, which is an internationally recognized fingerprint identification database.

2 simulation content and analysis

The method provided by the invention is adopted to carry out an experiment for extracting the direction field of the fingerprint image from the fingerprint image. Fig. 3 is a schematic diagram of the direction field of the fingerprint image, wherein the continuous black dotted line represents the extracted direction field of the fingerprint.

The method can quickly and accurately extract the direction field of the fingerprint image and reduce the influence of noise in the fingerprint image. Accurate direction information can also be obtained at the scar of the fingerprint image. The direction field extracted by the invention is used for enhancing the fingerprint image, so that the fracture parts of ridge lines in the fingerprint image can be effectively connected, the generation of pseudo-fine nodes can be greatly reduced, and the identification precision of the fingerprint is improved.

Claims (1)

1. a fingerprint image direction field extraction method based on frequency information is characterized by comprising the following steps:

Step one, inputting a fingerprint image I (x, y) with the size of H multiplied by W, wherein H represents the height of the fingerprint image I (x, y), W represents the width of the fingerprint image I (x, y), and (x, y) represents the coordinates of the fingerprint image I (x, y);

step two, carrying out blocking processing on the fingerprint image I (x, y), dividing the fingerprint image I (x, y) into image blocks with the size of m multiplied by n, overlapping two adjacent image blocks, wherein the width of an overlapping area is num pixels, and obtaining a fingerprint image block I_ij(x, y) where I denotes a fingerprint image block I_ij(x, y) is located in the ith row of the fingerprint image I (x, y), j represents the fingerprint image block I_ij(x, y) is located in the jth column of the fingerprint image I (x, y);

step three, calculating fingerprint image block I_ijThe average value avg of all pixel values in (x, y) constitutes a matrix V (x, y) of size m × nthe value of each point in V (x, y) is avg, and the matrix V (x, y) is used to image the fingerprint image block I_ij(x, y) removing the direct current component to obtain an image D (x, y) with the direct current component removed; performing Fourier transform on the image D (x, y) with the direct-current components removed to obtain a spectrogram F (x, y), and squaring the pixel value of each point in the spectrogram F (x, y) to obtain an energy map E (x, y);

Step four, all energy maximum value points are obtained in the energy diagram E (x, y), all the energy maximum value points are arranged in a descending order according to the energy values, and the energy maximum value points after the ordering are placed in a queue Q_kWhere k 1,2, N denotes the number of energy maxima in the energy diagram E (x, y);

step five, calculating a queue Q_kDirection o of the first two points_kAnd frequency f_kwhere k is 1,2, and setting a frequency threshold Th 1;

step six, if f₁Not more than the frequency threshold Th1, step seven is executed, otherwise, the fingerprint image block I is processed_ijDirection O of (x, y)₁The size of (i, j) is set to o₁fingerprint image block I_ijFrequency F of (x, y)₁The size of (i, j) is set to f₁Executing the step eight;

Step seven, if f₂Greater than the frequency threshold Th1, the fingerprint image block I_ijDirection O of (x, y)₁The size of (i, j) is set to o₂Fingerprint image block I_ijFrequency F of (x, y)₁the size of (i, j) is set to f₂Otherwise, the fingerprint image block I_ijThe direction and frequency of (x, y) are respectively set as constant a, a is used for representing fingerprint image block I_ijDirection O of (x, y)₁(i, j) and frequency F₁(i, j) cannot pass through the computation queue Q_kDirection o of the first two points_kAnd frequency f_kobtaining;

Step eight, repeating the step three to the step seven, and solving all fingerprint image blocks I_ijDirection O of (x, y)₁(i, j) and frequency F₁(i,j)；

step nine, fingerprint image block I_ijDirection O of (x, y)₁(i, j) and frequency F₁(I, j) carrying out first correction to obtain a fingerprint image block I_ijDirection O of (x, y)₂(i, j) and frequency F₂(i,j)；

Step ten, fingerprint image block I_ijdirection O of (x, y)₂(i, j) and frequency F₂(I, j) carrying out second correction to obtain a fingerprint image block I_ij(x, y) final direction O₃(i, j) and frequency F₃(i,j)。