CN102368333A - Fingerprint enhancement method based on time domain and frequency domain filtering - Google Patents

Abstract

The invention relates to a fingerprint enhancement method based on time domain and frequency domain filtering. The method comprises the following steps: a. carrying out enhancement processing on a time domain fingerprint image; b. carrying out frequency domain filtering enhancement processing simultaneously when carrying out enhancement processing on the time domain fingerprint image; c. carrying out accumulation on a time domain enhancement fingerprint image and a frequency domain enhancement fingerprint image according to each corresponding pixel point to realize fusion of pixel points, a fingerprint enhancement image is obtained, and image gray scale adjustment is carried out. According to the fingerprint enhancement method based on time domain and frequency domain filtering, through carrying out time domain filtering on a fingerprint image and carrying out filtering enhancement at a frequency domain simultaneously, an enhancement purpose is achieved, fingerprint lines are subjected to reparation effectively at a time domain, the fingerprint image is subjected to enhancement filtering from direction and frequency domain selectivity at the frequency domain, and the fingerprint image is substantially enhanced.

Description

Fingerprint Enhancement Method based on time domain and frequency domain filtering

Technical field

The present invention relates to the technical field of fingerprint image enhancement process, especially based on the fingerprint Enhancement Method of time domain and frequency domain filtering.

Background technology

How to differentiate accurately that a people's identity, the safety of protection information are critical technical problems that must solve the current information age.Traditional identity identifying method comprises identify label article (like key, certificate, atm card etc.) and identify label knowledge (like username and password etc.).But since mainly by all be external thing, in case the sign article of proof identity are stolen or forget with sign knowledge, its identity just is easy to pretended to be or replace by other people, thereby brings information security hidden danger.And the relatively more popular biological identification technology of research at present then is to utilize intrinsic physiological characteristic of human body and behavior act to carry out identification and checking like fingerprint recognition, recognition of face, speech recognition, iris recognition etc.These characteristics have uniqueness and stability, and constant throughout one's life, and these recognition methodss can overcome the deficiency of classic method greatly.Biological identification technology had obtained great concern in recent years in the world.The research report of international biological group is pointed out: than 26.84 hundred million dollars of 2007, the market of global biometrics identification technology in 2008 will reach 46.39 hundred million dollars, to 2012 the end of the year market will reach 74.56 hundred million dollars level.

Compare with other biological identification technology; Fingerprint recognition is in uniqueness, stability, accuracy of identification; Aspects such as anti-counterfeiting performance all have superiority; Become the safest, convenient at present, reliable identity authentication method gradually, its application in field of identity authentication has obtained extensive approval, has very strong authority of law property.Fingerprint image strengthens the first step as fingerprint recognition, has very crucial meaning.Because the quality of fingerprint recognition effect highly depends on the quality of fingerprint image, especially concerning based on the fingerprint recognition system of textural characteristics, the whether accurate recognition effect that will directly have influence on total system that RP is confirmed.The document proof fingerprint image after strengthen is than confirming RP more easily accurately from original image.In addition, for method for extracting fingerprint feature, generally be minutiae point (end points and bifurcation) through the image that takes the fingerprint based on minutiae point, carry out point-to-point coupling then and differentiate whether two pieces of fingerprints are identical fingerprints.In the gatherer process of fingerprint image,, the not equal reason of spot, fingerprint pressure connects noise owing to causing the fracture of fingerprint image ubiquity and the fork of collection.And the existence of these two kinds of noises all can effect characteristics point judgement, thereby generate the pseudo-characteristic point, and then influence the recognition effect of fingerprint, so these two kinds of noises all must elimination.If be used for the fingerprint image poor quality that minutiae point is extracted, what the minutiae point of extraction had will lose, and the reliability based on the matching process of minutiae point will reduce greatly like this.

Traditional fingerprint Enhancement Method mainly is divided into two kinds of time-domain filtering and frequency filterings.Time-domain filtering mainly through the structure suitable filters, carries out convolution algorithm in time domain to fingerprint image, thereby reaches the purpose of enhancing, like wave filter based on Gabor filter, and the nonlinear diffusion wave filter ^[6], the context wave filter ^[7]And method such as scale-space filtering.Be different from time domain filtering; The frequency domain Fourier transform can convert in frequency and does product calculation do convolution algorithm in time domain, thereby fast and effeciently image is handled, and reaches the purpose that strengthens image; As based on the direction frequency domain filter; The anisotropic wave filter, directly or the Short Time Fourier Transform analysis, and method such as " Log-Gabor " wave filter.But; These existing fingerprint Enhancement Method nearly all are to use single wave filter; And the parameter of selective filter is mostly based on the streakline structure of fingerprint; Because fingerprint image has the characteristic of non-stationary, the streakline structure of low-quality fingerprint image is very complicated especially, and the effect of such filtering reinforcement method often can not be satisfactory.

Summary of the invention

The technical matters that the present invention will solve is: in order to overcome the shortcoming of using single wave filter; A kind of fingerprint Enhancement Method based on time domain and frequency domain filtering is provided; Through fingerprint image being carried out the time-domain filtering of phase one; And filtered fingerprint image strengthened in the filtering that frequency domain carries out subordinate phase, thereby reach the enhancing purpose.This method can be repaired fingerprint ridge in time domain effectively, and on direction and frequency domain selectivity, fingerprint image is strengthened filtering respectively at frequency domain, thereby strengthens fingerprint image greatly.

The technical solution adopted for the present invention to solve the technical problems is: a kind of fingerprint Enhancement Method based on time domain and frequency domain filtering, and its concrete steps are following:

A. to time domain fingerprint image enhancement process: comprise topography's normalization, ask local streakline direction, compensation filter and smoothing processing; Time-domain filtering is mainly through the structure suitable filters; In time domain fingerprint image is carried out convolution algorithm, thereby reach the purpose of figure image intensifying, can extract the fact that produces very big influence to fingerprint characteristic to the streakline fracture and the streakline fuzzy region of fingerprint; Construct suitable time domain filtering; The convolution template of choose reasonable wave filter is carried out combing and repairing to these streakline zones, thereby reaches the purpose that strengthens fingerprint image.The selection of this filter parameter mainly is the parameter of structure convolution template;

B. in to time domain fingerprint image enhancement process, carry out the frequency domain filtering enhancement process: comprise the global image normalization, ask the adjustment of directional diagram, Gabor Filtering Processing and gradation of image; After time domain compensation filtering enhancing; The fingerprint image lines is comparatively clear; But in order to make fingerprint image in the application of reality, can extract characteristic more effectively, we have taked to strengthen based on the frequency domain of Gabor wave filter, and promptly the frequency domain fingerprint image strengthens.The frequency domain Fourier transform can convert in frequency and does product calculation do convolution algorithm in time domain, thereby fast and effeciently image is handled, and reaches the purpose that strengthens image.The frequency filter based on the Gabor wave filter of structure has good direction and frequency selectivity, can on direction and frequency domain selectivity, strengthen filtering to fingerprint image respectively at frequency domain;

C. strengthen fused images: the fingerprint image of time domain enhancing and the fingerprint image of frequency domain enhancing are added up according to each corresponding pixel, realize the fusion of pixel, obtain fingerprint and strengthen image and carry out the gradation of image adjustment.

Described enhancing fused images comprises the fusion and the gradation of image adjustment of pixel.

The beneficial effect of the fingerprint Enhancement Method based on time domain and frequency domain filtering of the present invention is: through fingerprint image is carried out time-domain filtering; Carrying out filtering at frequency domain simultaneously strengthens; Thereby reach the enhancing purpose; This method can be repaired fingerprint ridge in time domain effectively, and on direction and frequency domain selectivity, fingerprint image is strengthened filtering respectively at frequency domain, thereby strengthens fingerprint image greatly.

Description of drawings

Below in conjunction with accompanying drawing and embodiment the present invention is further specified.

Fig. 1 is the theoretical block diagram of implementation;

Fig. 2 is the theoretical block diagram of time-domain filtering of the present invention;

Fig. 3 is the former figure of fingerprint of time domain fingerprint image enhancing step 1 and the image after the local normalization;

Fig. 4 is the theoretical block diagram of frequency domain filtering of the present invention;

Fig. 5 is that time-domain and frequency-domain of the present invention strengthens the image co-registration block diagram;

Fig. 6 is that fingerprint image 1 strengthens experimental result;

Fig. 7 is that fingerprint image 2 strengthens experimental result.

(a) is the former figure of fingerprint of time domain fingerprint image enhancing step 1 among the figure; (b) be image after time domain fingerprint image enhancing step 1 local normalization; (c) be the former figure of fingerprint that fingerprint image 1 strengthens experiment; (d) be that the fingerprint that fingerprint image 1 strengthens behind the time-domain filtering of experiment strengthens image; (e) be that the fingerprint that fingerprint image 1 strengthens behind the frequency domain filtering of experiment strengthens image; (f) the fingerprint enhancing image after the time-domain filtering that strengthens experiment for fingerprint image 1 strengthens with frequency domain filtering simultaneously; (g) be the former figure of fingerprint that fingerprint image 2 strengthens experiment; (h) be that the fingerprint that fingerprint image 2 strengthens behind the time-domain filtering of experiment strengthens image; (i) be that the fingerprint that fingerprint image 2 strengthens behind the experiment frequency domain filterings strengthens image; (j) be the fingerprint image 2 fingerprint enhancing image after strengthening the experiment time-domain filterings and strengthening simultaneously with frequency domain filtering.

Embodiment

Combine accompanying drawing that the present invention is done further detailed explanation now.These accompanying drawings are the synoptic diagram of simplification, basic structure of the present invention only is described in a schematic way, so it only show the formation relevant with the present invention.

A kind of fingerprint Enhancement Method based on time domain and frequency domain filtering as shown in Figure 1, its concrete steps are following

One, the time domain fingerprint image strengthens

Streakline fracture and streakline fuzzy region to fingerprint can be extracted the fact that exerts an influence to fingerprint characteristic; Construct the convolution template of suitable time domain filtering and choose reasonable wave filter; Combing and repairing are carried out in these streakline zones; Can reach the purpose that strengthens fingerprint image, the selection of filter parameter mainly is a structure convolution template parameter, and concrete steps as shown in Figure 2 are following:

Step 1: topography's normalization

Consider that each local intensity profile of image is unbalanced, thus adopt local normalization to handle to image, definition expectation average M among this paper ₀=128, the expectation variance

Local window size W * W is (8 * 8).Topography normalization be to each pixel img of image (i j) averages and variance for the part at center, again to central point img (i j) standardizes, thereby obtains better effect, and key step is following:

(1) obtains that (i j) is the average M and the variance V of the topography at center with img;

(2) by following formula to central point img (i, j) processing of standardizing;

$\mathrm{norimg}(i,j)=M0+\frac{V0(\mathrm{img}(i,j)-M(i,j))}{V(i,j)},(i,j)\∈\mathrm{img}\∩V(i,j)\≠0$

norimg(i，j)＝M0+100*(img(i，j)-M(i，j)，(i，j)∈img∩V(i，j)＝0 (2)

(3) (i, j) edge value is handled to norimg.

if?norimg(i，j)＜0，norimg(i，j)＝0，

esls?if?norimg(i，j)＞255，norimg(i，j)＝255 (3)

Result after normalization is handled as stated above is shown in Fig. 3 (b).

Step 2: ask local streakline direction

Fingerprint image adopts gradient algorithm that the fingerprint local direction is estimated after normalization is handled.Fingerprint image after the normalization is divided into non-overlapping, the size of piece is W * W, and (i j) for center and mask operator carry out convolution, like the sobel operator, obtains the horizontal gradient component U of each pixel in the piece with the pixel of piece central point _X(u, v), VG (vertical gradient) component U _V(u, v) and whole horizontal gradient component U _XX(i, j), VG (vertical gradient) component U _XV(i, j), through the upwards horizontal gradient component and (4)～(6) the formula acquisition by formula of VG (vertical gradient) component of point of counterparty.

In order accurately to estimate the streakline direction; After to the horizontal gradient component and VG (vertical gradient) component of obtaining pixel; It is carried out gaussian filtering handle, then directional diagram is handled, then directional diagram is carried out smoothly with Gaussian filter; Carry out gaussian filtering at last again and handle, for next step compensation filter is got ready.

Step 3: compensation filter and smoothing processing

Repair to cut off crestal line and suppress non-crestal line information according to the half-tone information of the direction of crestal line and neighborhood and compensate filtering, smothing filtering is that low frequency strengthens the spatial domain filtering technique, and its purpose is a Fuzzy Processing and abate the noise.The smothing filtering of spatial domain adopts simple average method, promptly asks the average gray value of this pixel neighborhood pixels point.The size of neighborhood is directly related with level and smooth effect, and neighborhood is big more, and smooth effect is good more, but neighborhood is excessive, marginal information is lost in a large number, thereby output image is thickened, and therefore needs choose reasonable neighborhood size.By formula (7～9) compensate directional diagram:

i′＝i+mcos(oimg(i，j))+n*sin(oimg(i，j)) (8)

j′＝j-msin(oimg(i，j))+n*cos(oimg(i，j)) (9)

Wherein oimg (i is the local direction obtained in second step j), and wherein window size is w * h, w=4, h=14, the selection of these values is to come according to the empirical value of repeatedly experiment.

The frequency domain fingerprint image strengthens

After time domain compensation filtering enhancing, the fingerprint image lines is comparatively clear, but in order to make fingerprint image more effectively extract characteristic, can take to strengthen based on the frequency domain of Gabor wave filter, and promptly the frequency domain fingerprint image strengthens.The frequency domain Fourier transform can convert in frequency and does product calculation do convolution algorithm in time domain, thereby fast and effeciently image is handled, and reaches the purpose that strengthens image.The frequency filter based on the Gabor wave filter of structure has good direction and frequency selectivity, can on direction and frequency domain selectivity, strengthen filtering to fingerprint image respectively at frequency domain, and its concrete performing step as shown in Figure 4 is following:

Step 1: global image normalization

After the process enhancement process first time; Each local intensity profile of fingerprint image is relatively more balanced; Can adopt the global image normalization that each gray values of pixel points on the image is adjusted; Adjust to one to the contrast of different original images and average gray fixedly on the rank; Reduce the influence that different fingerprint image contrast differences are brought, definition expectation gray average expectation ash-Du variance

key step is following:

(1) gray-scale value of calculated fingerprint image and variance

$M=\frac{\underset{1}{\overset{i}{\mathrm{\Σ}}}\underset{1}{\overset{i}{\mathrm{\Σ}}}\mathrm{enhimg}(i,j)}{R*L},(i,j)\∈\mathrm{enhimg}---\left(10\right)$

$V=\frac{\underset{1}{\overset{i}{\mathrm{\Σ}}}\underset{1}{\overset{i}{\mathrm{\Σ}}}{(\mathrm{enhimg}(i,j)-M)}^2}{R*L},(i,j)\∈\mathrm{enhimg}---\left(11\right)$

R, L be the row value and the train value of presentation video respectively, and enhimg is the image information after strengthening for the first time.

(2) by following formula image is standardized

Step 2: ask directional diagram

In the image on the different directions point corresponding gray scale be vicissitudinous, the gray scale in crestal line differs very little, and is maximum but the gray scale on the vertical direction differs, and can obtain directional diagram through compute gradient, key step is following:

(1) utilizes sobel to calculate each pixel and do two convolution, obtain the VG (vertical gradient) component U of each pixel _V(i is j) with horizontal gradient component U _X(i, j);

(2) the direction expansion with each pixel gradient vector is twice;

(3) ask the average gradient at each pixel place vectorial, local field size is W ₁* W ₁. (W ₁=15), formula is:

Then

that obtains can not express all directions of fingerprint, and the angle of direction by formula (16) is adjusted:

Wherein (i j) is adjusted angle to O.

Step 3:Gabor Filtering Processing

The Gabor wave filter has good direction and frequency selectivity, can when preserving correct crestal line and valley line structure, remove noise with it as frequency filter, and this paper adopts the Gabor wave filter to realize the enhancing of fingerprint image.Definition W ₂=10 are filtering mask size, U _X=3 are the logical size of frequency band, U _VBe the logical size of direction band, ∫=0.1 is the streakline average frequency, and key step is following:

(1) utilizes the gabor wave filter to be used in reference to print image and handle, strengthen template antithetical phrase piece with formula (17) as sub-piece and strengthen;

$E(i,j)=\exp (-\frac{1}{2}(\frac{x^2}{{\mathrm{Ox}}^2}+\frac{y^2}{{\mathrm{Oy}}^2}))*\cos (2*\mathrm{\π}*f*x)$

x＝v*sin(O(i，j))+u*cos(O(i，j))

y＝v*cos(O(i，j))-u*sin(O(i，j)) (17)

(2) the sub-piece after will strengthening by formula (18) be merged into the complete finger print image;

Wherein E is the image after strengthening, and the Gabor wave filter can reasonably carry out filtering to fingerprint image at each topography's piece, improves signal noise ratio (snr) of image, thereby can preserve comparatively complete sum fingerprint ridge information clearly.

Strengthen image co-registration

In conjunction with like Fig. 5, strengthen image co-registration and may further comprise the steps:

Step 1: the fusion of pixel;

Time domain enhancing fingerprint image and frequency domain enhancing fingerprint image are added up according to each corresponding pixel, realize the fusion of pixel.

Step 2: gradation of image adjustment;

The enhancing image that merges is at last carried out the gray scale adjustment, and key step is following:

(A) determine the maximum enhanced fingerprint image gray value

and the minimum gray value

and then take the difference

(2) each pixel with E deducts minimum value;

(3) utilize formula (20) that each gray values of pixel points is adjusted.

E2(i，j)＝E(i，j)*256/val (20)

The fingerprint Enhancement Method based on time domain and frequency domain filtering of invention; Adopt the Matlab programming language on the platform of matlab7.0, to realize; Chosen two fingerprint images and done experiment test, experimental result such as Fig. 6 and shown in Figure 7 can find out from experimental result; For different images; The effect that different filtering modes are obtained is different, Fig. 6 and Fig. 7 former figure before the effect after out-of-date, frequency domain strengthen is superior to strengthening, but two pieces of fingerprint images are best passing through the reinforced effects that obtains after time-domain filtering and the frequency domain filtering enhancement process.Reason is: time-domain filtering is handled the contrast of the crestal line and the valley line that can improve fingerprint effectively; And along the crestal line of combing of fingerprint local direction and compensation fingerprint image, particularly streakline fracture and streakline fuzzy region, frequency domain filtering is to utilize frequency bandpass filter on frequency domain, to carry out fingerprint image to strengthen; It has good direction and selects and frequency selective characteristic; Can make full use of the frequency and the directional information of streakline in the regional area, outstanding streakline inherent structure is removed picture noise; And can fingerprint ridge line and the valley line structure is distortionless remain, reach the purpose that strengthens fingerprint image.Fig. 6 (f) and Fig. 7 (j) are the images of handling through time-domain and frequency-domain, combine the advantage of time-domain filtering and frequency domain filtering, and therefore, the reinforced effects of two images is best.

With above-mentioned foundation desirable embodiment of the present invention is enlightenment, and through above-mentioned description, the related work personnel can carry out various change and modification fully in the scope that does not depart from this invention technological thought.The technical scope of this invention is not limited to the content on the instructions, must confirm its technical scope according to the claim scope.

Claims (2)

1. fingerprint Enhancement Method based on time domain and frequency domain filtering, it is characterized in that: its concrete steps are following:

A. to time domain fingerprint image enhancement process: comprise topography's normalization, ask local streakline direction, compensation filter and smoothing processing;

B. in to time domain fingerprint image enhancement process, carry out the frequency domain filtering enhancement process: comprise the global image normalization, ask the adjustment of directional diagram, Gabor Filtering Processing and gradation of image;

C. strengthen fused images: the fingerprint image of time domain enhancing and the fingerprint image of frequency domain enhancing are added up according to each corresponding pixel, realize the fusion of pixel, obtain fingerprint and strengthen image and carry out the gradation of image adjustment.

2. the fingerprint Enhancement Method based on time domain and frequency domain filtering according to claim 1 is characterized in that: described enhancing fused images comprises the fusion and the gradation of image adjustment of pixel.

Images