CN106342325B - A kind of region segmentation method of fingerprint image - Google Patents

Abstract

The region segmentation method that the present invention relates to a kind of fingerprint image, is characterized in that: first fingerprint image is carried out to normalization process, press the inhomogeneous impact on the fingerprint image collecting to solve; Then, utilize 3 × 3 mask methods to judge crestal line direction, and taking the gray scale difference value of crestal line direction and perpendicular direction as according to carrying out the pre-segmentation in region; Finally, pre-segmentation result is carried out to smoothing processing and opening operation, on original fingerprint image, extract texture region, complete cutting apart of texture region and non-texture region. Utilize the inventive method, can realize effectively and the fast Region Segmentation of fingerprint image, and can remove well the non-texture region such as frame and noise background of fingerprint capturer, cut zone is accurately smooth, can effectively improve the performance of automatic system of fingerprint recognition.

Description

A kind of region segmentation method of fingerprint image

Technical field

The present invention relates to a kind of region segmentation method of fingerprint image, belong to Digital Image Processing and area of pattern recognition,Be suitable for the finger-print region cutting procedure in computer Automated Fingerprint Identification System.

Background technology

Fingerprint recognition is a kind of important biological characteristic authentication technique, and along with the development of computer technology, fingerprint is knownThe application of other technology is also more and more wider. In automatic system of fingerprint recognition, the fingerprint image that collecting device obtains is oneThe image that width contains more noise, must, through pretreatment, remove a large amount of noise signals, obtains a width streakline clearPoint and line chart, just can carry out extraction and the coupling of fingerprint characteristic. Fingerprint Image Segmentation is extremely important in image pretreatmentA step, be usually located at pretreated front end, its objective is poor quality in a fingerprint image, in subsequent treatment veryDifficult image-region and the texture region recovering makes a distinction, and makes subsequent treatment can concentrate on texture region. It is not only wantedAsk and remove as much as possible non-texture region, also will as far as possible intactly retain texture region. After Fingerprint Image Segmentation, can keep awayExempt to extract feature in noise and background area, improve the reliability of feature extraction, save the processing time, to carrying simultaneouslyHigh whole system performance has great significance.

Traditional Fingerprint Image Segmentation is mainly divided into (the side of cutting apart based on gradation of image characteristic by cutting apart decision conditionPoor method) and utilize cut apart (direction method) that image orientation information carries out. Variance method splitting speed is very fast, but it is only at non-lineReason region effect is better, to the crestal line in texture region and valley line None-identified. Direction method is utilized the directional information of fingerprintCut apart, can obtain desirable segmentation effect at texture region, but it is in non-texture region and the poor district of texture qualityTerritory effect declines.

Traditional Fingerprint Image Segmentation is according to the concrete form of cutting apart, can be divided into block-basedly cut apart, based on pixelCutting apart of point and cutting apart of the outline that takes the fingerprint. Wherein block-based method is the most common, and these class methods conventionally willFingerprint image is divided into fixed-size, determines that according to the characteristic of every this piece is texture region or non-texture region.Be a kind of special circumstances for the moment based on being cutting apart of pixel block size in fact, its speed is cut apart slowly than piece, but cuts apartRear image border is smooth.

Said method respectively has pluses and minuses, is difficult to accomplish not only accurately but also rapid.

Summary of the invention

The technical problem solving

For fear of the deficiencies in the prior art part, the present invention proposes a kind of fingerprint image region segmentation method efficiently.

Thought of the present invention is: gray scale and the directional information of comprehensive utilization fingerprint image, and according to orthogonal direction gray valueVariation, judge texture region, thereby complete the texture region of fingerprint image and cutting apart of non-texture region.

Technical scheme

A region segmentation method for fingerprint image, is characterized in that comprising the following steps:

Step 1 normalization: utilizeTo original fingerprint imageCarry out normalization process, make the average gray of original fingerprint image and variance all reach expectationValue;

Wherein: I (i, j) represents to be positioned at the grey scale pixel value of the capable j row of original fingerprint image i;TableShow the grey scale pixel value that is positioned at the capable j row of fingerprint image i after normalization process; M₀Represent to expectAverage gray, M₀＞0；VAR₀Represent the variance of expecting, VAR₀> 0; M representsThe average gray of original fingerprint image; VAR represents the variance of original fingerprint image;

Step 2 region pre-segmentation: utilizing block size is the ridge that 3 × 3 mask means calculates fingerprint image after normalization processLine direction iDir and vertical direction iDir thereof⁺, utilize in 3 × 3 pixel on this both directionThe difference of gray value judges whether the central pixel point of this piece is texture region, obtains region pre-Cut apart template segtemp, concrete steps are as follows:

Step a: by the fingerprint image after normalization processBe divided into 3 × 3 can overlapping fritter, 3In × 3, in calculated direction 1, direction 2, direction 3 and direction 4, remove central pixel point respectivelyAverage gray Gmean (1), the Gmean (2) of two pixels in addition, Gmean (3), Gmean(4)；

Described direction 1 represents horizontal direction; Described direction 2 represents positive 45 degree directions; DescribedDirection 3 represent vertical direction; Described direction 4 represents negative 45 degree directions;

Step b: utilize Gdiff (1)=| Gman (1)-Gman (3) | in calculated direction 1 except central pixel pointIn the average gray Gmean (1) of two pixels in addition and direction 3, remove central pixel pointThe absolute value Gdiff (1) of the difference of the average gray Gmean (3) of two pixels in addition; ProfitWith Gdiff (2)=| Gman (2)-Gman (4) | in calculated direction 2 except central pixel point twoIn the average gray Gmean (2) of individual pixel and direction 4 except central pixel point twoThe absolute value Gdiff (2) of the difference of the average gray Gmean (4) of individual pixel;

Step c: in the time of Gdiff (1) > Gdiff (2), iMax=1, determines preliminary crestal line direction iMaxFor direction 1; In the time of Gdiff (1)≤Gdiff (2), iMax=2, determines preliminary crestal line direction iMaxFor direction 2;

Steps d: pressMeterCalculate crestal line direction iDir; UtilizeCalculate crestal line Vertical SquareTo iDir⁺；

Wherein:The gray value of the central point pixel in representing 3 × 3, in normalization processAfter fingerprint image in position be i₀Row j₀Row;

Step e: calculate upper two pictures except central point pixel of crestal line direction iDir in 3 × 3Average gray Gmean (iDir) and the iDir of element⁺In direction except central point pixel twoAverage gray Gmean (the iDir of pixel⁺)；

Step f: utilizeObtainPre-segmentation template segtemp;

Wherein: segtem (i, j) represents to be positioned at the pixel value of the capable j row of the upper i of pre-segmentation template segtemp,The size of pre-segmentation template segtemp is identical with the size of original fingerprint image; || represent to get absolutelyTo value; t₀Represent segmentation threshold, t₀＞0；

Step 3 segment smoothing: utilize size pre-segmentation template segtemp to be carried out smoothly for the piece of m × n, concrete steps asUnder:

Step a: in statistics pre-segmentation template segtemp centered by each pixel segtem (i, j)The pixel number Sum that the interior pixel value of piece that size is m × n is 0;

Step b: utilizeCutting apart after obtaining smoothlyTemplate segtemp1;

Wherein: segtemp1 (i, j) represents to be positioned at cuts apart the capable j row of the upper i of template segtemp1 after level and smoothPixel value;

Step 4 opening operation: the template segtemp1 of cutting apart after level and smooth is carried out to opening operation, obtain final Region Segmentation template segtemp2；

Step 5 texture region is cut apart: utilize I_wen=(1-segtemp2) I carries out texture region and non-to original fingerprint imageFinally cutting apart of texture region;

Wherein: I_wenFor final Region Segmentation image; Represent dot-product operation;

Wherein, the size of the piece in segment smoothing meets: m=n, and m and n are the even number in (0,64).

Beneficial effect

The region segmentation method of the fingerprint image that the present invention proposes, has fully utilized directional information and half-tone information, in conjunction withVariance method splitting speed fast, good and direction method is good at texture region segmentation effect at non-texture region segmentation effectAdvantage, cut zone is smooth, and segmentation precision is high. In 3 × 3, calculate crestal line direction and cut apart, having accelerated fortuneCalculation speed can be applied this algorithm in engineering.

Brief description of the drawings

Fig. 1: the basic flow sheet of the inventive method;

Fig. 2: optical collector image region segmentation process schematic diagram;

A) original fingerprint image;

B) fingerprint image after normalization process;

C) pre-segmentation result images;

D) level and smooth result images;

E) image after opening operation;

F) the final area segmentation result image of original fingerprint image;

Fig. 3: conventional method and the inventive method are cut apart template contrast;

A) paper stamp original fingerprint image;

B) conventional method is cut apart template;

C) the inventive method is cut apart template;

Detailed description of the invention

Now in conjunction with the embodiments, the invention will be further described for accompanying drawing:

For the hardware environment implemented be: ACPI Multiprocessor PC-4400+ computer, 1.00GB internal memory,128M video card; The software environment of operation is: Window XP, has realized with MATLAB 7.0.1 programming languageThe method that the present invention proposes.

1. pair original fingerprint image carries out normalization process

It is to make the average gray of image and variance all reach predefined that fingerprint image is carried out to normalized objectAverage gray and the variance expected, strengthen integral image contrast. Normalization process can effectively solve and press notThe evenly impact on the fingerprint image collecting. Normalization adopts following formula to realize:

$\tilde{I}(i,j)=\left\{\begin{array}{cc}{M}_0+\sqrt{{\mathrm{VAR}}_0{\left(I\right(i,j)-M)}^2/VAR}& I(i,j)\&GreaterEqual;M\\ M_0-\sqrt{{\mathrm{VAR}}_0{\left(I\right(i,j)-M)}^2/VAR}& I(i,j)<M\end{array}\right.$

Wherein, I (i, j) represents to be arranged in the grey scale pixel value of the capable j row of original fingerprint image I i;Represent to be just positioned atRuleization are processed rear fingerprint imageThe grey scale pixel value of the capable j row of middle i; M, VAR is respectively the gray scale of original fingerprint imageMean value and variance, M₀、VAR₀For average gray and the variance expected,, get M here₀＝120，VAR₀＝120。

2. pair normalization fingerprint image after treatment carries out pre-segmentation

By the fingerprint image after normalization processBe divided into 3 × 3 can overlapping block, suppose that any pixel is9 pixels in 3 × 3 centered by this pixel are respectively:

$\tilde{I}(i-1,j-1),\tilde{I}(i-1,j),\tilde{I}(i-1,j+1)$

$\tilde{I}(i,j-1),\tilde{I}(i,j),\tilde{I}(i,j+1)$

$\tilde{I}(i+1,j-1),\tilde{I}(i+1,j),\tilde{I}(i+1,j+1)$

Calculate respectively 3 × 3 interior levels, positive 45 degree, vertically and in negative 45 degree directions except central pixel point twoThe average gray of individual pixel, the code name of establishing this four direction is respectively 1,2,3,4, has:

Horizontal direction

Positive 45 degree directions

Vertical direction

Negative 45 degree directions

By the average gray in the perpendicular direction of average gray Gmean obtained above (1), Gmean (2) differenceGmean (3), Gmean (4) subtract each other and take absolute value, and formula is as follows:

Gdiff (1)=| Gmean (1)-Gmean (3) | level and vertical direction average gray poor

Gdiff (2)=| Gmean (2)-Gmean (4) | positive and negative 45 spend the poor of direction average gray

Get a prescription that the absolute value of difference is larger to being possible crestal line direction, if i.e. Gdiff (1) > Gdiff (2),The possible crestal line direction in this pixel place is 1 or 3; If Gdiff (1) < Gdiff (2), this possible crestal line side of pixel placeTo being 2 or 4. If possible crestal line direction is iMax, tentatively judge the direction of crestal line with following formula:

$iMax=\left\{\begin{array}{cc}1& Gdiff\left(1\right)>Gdiff\left(2\right)\\ 2& Gdiff\left(1\right)\&le;Gdiff\left(2\right)\end{array}\right.$

For central point pixelThe direction of determining the average gray less with its gray value difference is crestal line directionIDir, formula is as follows:

Determine that after the crestal line direction iDir at this pixel place,, if fingerprint ridge line direction is 1 or 2, it is verticalDirection is correspondence 3 or 4 respectively, utilizes following formula can obtain crestal line vertical direction iDir⁺：

${\mathrm{iDir}}^{+}=\left\{\begin{array}{cc}iDir+2& 1\&le;iDir\&le;2\\ iDir-2& 3\&le;iDir\&le;4\end{array}\right.$

Calculate this crestal line direction iDir and its orthogonal direction iDir⁺On the absolute value of difference of gray scale |Gmean(iDir)-Gmean(iDir⁺) |, if its result is less than segmentation threshold t₀, the gray scale difference of two direction is describedDifferent little, think that this point is non-texture region point, should remove, be zero pre-segmentation template segtemp entirely at initial valueThe middle value corresponding points is set to 1, otherwise is set to zero. Formula is as follows:

Wherein, the size of pre-segmentation template segtemp is identical with the size of original fingerprint image; Segtemp (i, j) represents positionThe pixel value that the capable j of the upper i of pre-segmentation template segtemp that is zero entirely in initial value is listed as; Here get t,₀＝15。

3. pair pre-segmentation template is carried out smoothly

After above-mentioned processing, obtain a width two-value pre-segmentation image template segtemp, next segtemp is carried outSmoothing processing.

Size of model all 1's matrix segtemp1 identical with pre-segmentation template segtemp, then by pre-segmentationTemplate segtemp be divided into size be 16 × 16 can be overlapping piece, and ask for pixel value sum Sum in piece, whenWhen Sum > 128, illustrate that the non-texture region point in this piece is many, think that the central pixel point in this piece is also non-texture areaTerritory point, establishing this central pixel point is segtemp (i, j), some segtemp1 (i, j) corresponding in matrix segtemp1 is set to0. The piece that is m × n to all sizes all calculates by said process and judges, completes locating matrix segtemp1Reason.

Like this, the segtemp1 obtaining is the image template of cutting apart after level and smooth, and in segtemp1, the part that value is 1 isThe texture region of fingerprint image, the part that value is 0 is the non-texture region of fingerprint image.

4. pair level and smooth result is carried out opening operation

The texture region (part that value is 1) of cutting apart in image template segtemp1 after level and smooth is corroded with swollenSwollen, the non-texture regions such as the frame of elimination aperture and collector. In matlab, adopt the disk that radius is δ=20Corrode and expand, namely opening operation, utilizes opening operation can eliminate wisp, at very thin some place separatorBody, level and smooth larger object border, but the area of while the original object of not obvious change. So just obtain final dividingCut template segtemp2, the point that segtemp2 intermediate value is 1 represents non-texture region, and the point that value is 0 represents texture region.

5. fingerprint image texture region is cut apart

Utilize the template segtemp2 of cutting apart in step 4 to carry out Region Segmentation to original fingerprint image. If cut apart templateIn segtemp2, the value of pixel is 1, and explanation is non-texture region, on original fingerprint image by the picture of corresponding point positionElement value is set to 0, is 0 if cut apart the value of pixel in template segtemp2, and explanation is texture region, at original fingerprintOn image, do not process, so just completed finally cutting apart of texture region and non-texture region.

From the large storehouse of public security fingerprint, randomly draw 6000 pieces of fingerprints, wherein the fingerprint image of stamp scanning and collector rollScan image respectively accounts for half. The fingerprint image that utilizes in addition collector to gather 1000 pieces of front stamps is cut apart examinationTest, Fig. 2 is optical collector image texture Region Segmentation process schematic diagram, and Fig. 3 is conventional method and the inventive methodCut apart template contrast. The result of cutting apart confirms, the inventive method comprehensive utilization direction and half-tone information, more traditional point countingCut region smooth, segmentation precision is high, has overcome conventional method texture region and has had a lot of stairstepping blocks, easily leadsCause inside cut apart excessive, lose characteristic point or edge and cut apart the deficiency that has error, effectively avoid in noise and backgroundExtract minutiae in region, has improved the reliability of feature extraction, saves the processing time, can in engineering, apply.

Claims (2)

1. a region segmentation method for fingerprint image, is characterized in that comprising the following steps:

Step 1 normalization: utilizeTo original fingerprint imageCarry out normalization process, make the average gray of original fingerprint image and variance all reach expectationValue;

Wherein: I (i, j) represents to be positioned at the grey scale pixel value of the capable j row of original fingerprint image i;TableShow the grey scale pixel value that is positioned at the capable j row of fingerprint image i after normalization process; M₀Represent to expectAverage gray, M₀＞0；VAR₀Represent the variance of expecting, VAR₀> 0; M representsThe average gray of original fingerprint image; VAR represents the variance of original fingerprint image;

Step 2 region pre-segmentation: utilizing block size is the ridge that 3 × 3 mask means calculates fingerprint image after normalization processLine direction iDir and vertical direction iDir thereof⁺, utilize in 3 × 3 pixel on this both directionThe difference of gray value judges whether the central pixel point of this piece is texture region, obtains region pre-Cut apart template segtemp, concrete steps are as follows:

Step a: by the fingerprint image after normalization processBe divided into 3 × 3 can overlapping fritter, 3In × 3, in calculated direction 1, direction 2, direction 3 and direction 4, remove central pixel point respectivelyAverage gray Gmean (1), the Gmean (2) of two pixels in addition, Gmean (3), Gmean(4)；

Described direction 1 represents horizontal direction; Described direction 2 represents positive 45 degree directions; DescribedDirection 3 represent vertical direction; Described direction 4 represents negative 45 degree directions;

Step b: utilize Gdiff (1)=| Gman (1)-Gman (3) | in calculated direction 1 except central pixel pointIn the average gray Gmean (1) of two pixels in addition and direction 3, remove central pixel pointThe absolute value Gdiff (1) of the difference of the average gray Gmean (3) of two pixels in addition; ProfitWith Gdiff (2)=| Gman (2)-Gman (4) | in calculated direction 2 except central pixel point twoIn the average gray Gmean (2) of individual pixel and direction 4 except central pixel point twoThe absolute value Gdiff (2) of the difference of the average gray Gmean (4) of individual pixel;

Step c: in the time of Gdiff (1) > Gdiff (2), iMax=1, determines preliminary crestal line direction iMaxFor direction 1; In the time of Gdiff (1)≤Gdiff (2), iMax=2, determines preliminary crestal line direction iMaxFor direction 2;

Steps d: pressMeterCalculate crestal line direction iDir; UtilizeCalculate crestal line Vertical SquareTo iDir⁺；

Wherein:The gray value of the central point pixel in representing 3 × 3, in normalization processAfter fingerprint image in position be i₀Row j₀Row;

Step e: calculate upper two pictures except central point pixel of crestal line direction iDir in 3 × 3Average gray Gmean (iDir) and the iDir of element⁺In direction except central point pixel twoAverage gray Gmean (the iDir of pixel⁺)；

Step f: utilizeObtainPre-segmentation template segtemp;

Wherein: segtem (i, j) represents to be positioned at the pixel value of the capable j row of the upper i of pre-segmentation template segtemp,The size of pre-segmentation template segtemp is identical with the size of original fingerprint image; || represent to get absolutelyTo value; t₀Represent segmentation threshold, t₀＞0；

Step 3 segment smoothing: utilize size pre-segmentation template segtemp to be carried out smoothly for the piece of m × n, concrete steps asUnder:

Step a: in statistics pre-segmentation template segtemp centered by each pixel segtem (i, j)The pixel number Sum that the interior pixel value of piece that size is m × n is 0;

Step b: utilizeCutting apart after obtaining smoothlyTemplate segtemp1;

Wherein: segtemp1 (i, j) represents to be positioned at cuts apart the capable j row of the upper i of template segtemp1 after level and smoothPixel value;

Step 4 opening operation: the template segtemp1 of cutting apart after level and smooth is carried out to opening operation, obtain final Region Segmentation template segtemp2；

Step 5 texture region is cut apart: utilize I_wen=(1-segtemp2) I carries out texture region and non-to original fingerprint imageFinally cutting apart of texture region;

Wherein: I_wenFor final Region Segmentation image; Represent dot-product operation.

2. the region segmentation method of a kind of fingerprint image according to claim 1, is characterized in that: described region is flatThe size of the piece in cunning meets: m=n, and m and n are the even number in (0,64).