CN103400134B - Non-contact type finger vein sample area-of-interest exacting method - Google Patents

Abstract

The present invention is to provide non-contact type finger vein sample area-of-interest exacting method.Carry out Threshold segmentation by column by varimax finger vein image, carry out denoising by the method for connected domain area and searching finger tip；The upper and lower profile point of finger is utilized to simulate finger center line, using the slope of finger center line as the foundation of rotation correction；Utilize the feature that finger tip contour approximation is circle to ask for the effective contour point set of circular arc, with these effective contour point matching finger tips circle, obtain finger tip diameter of a circle and carry out area-of-interest location；The transverse width of area-of-interest be fixed width, longitudinally wide by finger contours lower edges internal tangent determine；Finally the area-of-interest obtained is carried out size normalization.The present invention takes into full account non-contact type finger venous collection process and feature of image, provides new solution for finger vena region of interesting extraction, utilizes finger center line to carry out rotation correction and finger tip circle positions, have the strongest anti-rotation, anti-translation.

Description

Non-contact type finger vein sample area-of-interest exacting method

Technical field

The present invention relates to a kind of mode identification technology, specifically a kind of finger vena sample area-of-interest carries Access method.

Background technology

Existing finger vein identification method great majority, based on complete finger-image information, directly process with whole finger Not only needing bigger memory space, operand is relatively big, and feature is not concentrated, and discrimination is relatively low.Contrast finger tip and finger root, hands Vein blood vessel ratio in the middle of referring to is more visible, and discrimination is big, is more beneficial for classifying as the object of feature extraction with this subregion. Therefore, select this part as our vein area-of-interest (Region Of Interest, ROI).

Apply for be 201210051702.3 patent document in, disclose one " finger vena based on rotation correction Interesting image regions extracting method ", utilize finger centroid to carry out image rotation correction, and vertical according to each column pixel in image Projection value on direction, determines finger the first joint position, carries out ROI location.The method has strict demand to picture quality, right Have the most bimodal after the projection of high-quality vein image column direction at two finger-joints；To low-quality vein image, After row projection, peak of curve is inconspicuous, or owing to illumination reason presents multimodal situation, it is impossible to realize being accurately positioned of ROI.

Summary of the invention

It is an object of the invention to provide one and can overcome skew, rotational differential present in gatherer process, wrapped Non-contact type finger vein sample area-of-interest exacting method containing the ROI region of abundant, stable feature.

The object of the present invention is achieved like this:

Determine including Threshold segmentation by column, image denoising, rotation correction, anchor point, region of interesting extraction and size are returned One changes, and uses varimax finger vein image to carry out Threshold segmentation by column, and by connected domain area and searching finger tip Method carry out denoising, be partitioned into finger areas；The upper and lower profile point of finger is utilized to simulate finger center line, with described finger center line Slope as the foundation of rotation correction；Utilize the feature that finger tip contour approximation is circle to ask for the effective contour point set of circular arc, enter One step these effective contour point matching finger tips circle, obtains finger tip diameter of a circle, carries out area-of-interest location with this；Interested The transverse width in region be fixed width, longitudinally wide by finger contours lower edges internal tangent determine；Finally to the sense obtained Interest region carries out size normalization.

Described employing varimax finger vein image carry out by column Threshold segmentation particularly as follows: based on image acquisition mistake In journey, the illumination of each row is equally distributed, using every string of image as a unit, uses varimax, calculates this The segmentation threshold of row, and carry out the global threshold segmentation of these row.

The described method by connected domain area and searching finger tip carries out denoising particularly as follows: calculate respectively according to zone marker The area of individual connected domain, retains the area in largest connected territory, eliminates little block distortion；For the noise of tip portion, according to Projection value on each column pixel vertical direction in image, determines the position of finger tip, eliminates the noise before finger tip.

Described rotation correction is particularly as follows: the point on finger center line is obtained by the midpoint of the upper and lower profile point of finger, by these points Simulating a near linear, the slope of straight line is positive and negative relevant with direction of rotation, and the size of slope is relevant with the anglec of rotation.

Described anchor point determines particularly as follows: utilize the feature that finger tip contour approximation is circle to ask in finger tip radius of circle Effective contour point set, further with these effective contour points by least square fitting finger tip circle, obtains finger tip diameter of a circle, On finger center line, the point away from one diametral distance of finger tip is for area-of-interest anchor point.

Described region of interesting extraction is particularly as follows: part secant on the left side from the beginning of anchor point, and it is fixing for parting secant on the right side away from anchor point Width；Cut-off rule is determined by the upper and lower edge internal tangent of finger contours folded by the cut-off rule of left and right up and down.

Described size normalization is particularly as follows: according to bilinear interpolation algorithm, carry out length and width by the area-of-interest of extraction and return One change processes, it is achieved obtain different samples is all the region that size is identical.

Because the collection of finger vena is contactless, not having the constraint of gim peg, finger is put naturally, same person Between different samples the most variant in horizontal offset, the anglec of rotation, these factors have impact on accuracy of identification.Therefore, first will Finger-image rotates to level, it is considered to finger gathers every time and offsets, and the finger length obtained is different, and therefore length information can not As the foundation of location, find the point that can carry out positioning on finger.The present invention is directed to the feature of contactless collection image, carry Go out to utilize finger center line to carry out rotation correction and method that finger tip circular diameter carries out positioning, it is possible to extract finger vein sense exactly Interest region, solves rotation, the impact of translation factor.

Present invention is characterized in that and take into full account non-contact type finger venous collection process and feature of image, quiet for finger Arteries and veins region of interesting extraction provides new solution, utilizes finger center line to carry out rotation correction and finger tip circle positions, tool There are the strongest anti-rotation, anti-translation.

Accompanying drawing explanation

The module composition of Fig. 1 non-contact type finger vein sample area-of-interest exacting method；

Fig. 2 (a)-Fig. 2 (b) is Threshold segmentation by column；Wherein Fig. 2 (a) original image, after Fig. 2 (b) Threshold segmentation by column Image；

Fig. 3 (a)-Fig. 3 (d) denoising process, wherein Fig. 3 (a) is the image after utilizing connected domain area to remove block distortion, Curve after the row projection of Fig. 3 (b) vein image object pixel, Fig. 3 (c) is the image removing finger tip front-end noise, and Fig. 3 (d) carries The finger areas taken；

Fig. 4 (a)-4 (b) rotation correction process, wherein Fig. 4 (a) is that the point on matching finger center line obtains finger center line, figure Finger vein image after 4 (b) rotation correction；

Fig. 5 (a)-Fig. 5 (b) anchor point determines process, wherein Fig. 5 (a) rough estimate finger tip radius, obtains effective contour point, Fig. 5 (b) circle matching determines finger tip arc diameter；

Fig. 6 (a)-Fig. 6 (b) region of interesting extraction process, wherein Fig. 6 (a) area-of-interest size determines schematic diagram, figure 6 (b) finger vena area-of-interest；

Fig. 7 size normalized vein area-of-interest.

Detailed description of the invention

Detailed description of the invention the following detailed description of the present invention.

1, Threshold segmentation by column

The original image size collected is 320 × 240, from Fig. 2 (a) it can be seen that background area accounts for image major part Ratio, directly processes not only increase amount of calculation by entire image and takies memory space, and be unfavorable for identifying.It is thus desirable to first carry Take finger areas, reduce background area interference.The thresholding operation of image can realize separating background and target object.

Threshold operation is the grey value difference according to different pixels, and the pixel that image is more than a certain threshold value is placed on one Rise, form target area；Putting together less than the pixel of this threshold value, form background area, then gray level image becomes only two The bianry image of individual grade.

Threshold segmentation is formulated and is exactly

$g(x,y)=\left\{\begin{array}{c}1,f(x,y)\&GreaterEqual;t\\ 0,f(x,y)<t\end{array}\right.---\left(1\right)$

In formula, (x, y) is image intensity value to f, and t is segmentation threshold.Divide an image into two parts with t for boundary, target is with 1 Representing, background represents with 0.

Varimax is a kind of adaptive threshold choosing method, is analyzed from the generic attribute in region, after making segmentation Target is maximum with background between group variable, and t now is optimal threshold.First the image that intensity profile is m level is obtained each gray scale The probability of value, is then classified as two groups with variable T, calculates each group of probability and between group variable,

δ²(T)=w₀(μ₀-μ)²+w₁(μ1-μ)² (2)

In formula, μ₀、μ₁It is the average gray of two groups, w₀、w₁Being the probability of happening of two groups, μ is that entire image gray scale is average Value.From changing T between 0～m-1, as between group variable δ²T time maximum, is threshold value.

Research finds, in image acquisition process, every intensity of illumination suffered by string pixel is uniformly distributed, and is identical, target Easily divide with background, it is contemplated that realized the acquisition of finger areas by the operation of maximum variance thresholding by column.

Concrete thought is: every string of image is regarded as a unit, uses varimax, calculates the segmentation of these row Threshold value, and carry out the global threshold segmentation of these row.Although this algorithm adds segmentation times, but owing to participating in computing every time Pixel is few, and computing is the most uncomplicated, fast.Through separation target by column and background, it is possible to obtain comprising completely The bianry image of target object.

The bianry image that Fig. 2 (b) obtains for using Threshold segmentation by column, it is possible to be partitioned into complete finger areas.

2, image denoising

Due to light uneven in gatherer process, result in thresholding and process and do not reach preferable effect, though Right whole finger areas to be identified occurs in that, but meanwhile the impurity of some bulks also occurs in that.Complete in order to extract Finger areas, it is necessary to the noise outside finger areas is completely removed.It is observed that the area of the block impurity of major part is general all Less than finger areas, if setting an area threshold, remove the extrinsic region less than this value, then major part extrinsic region Will be removed.

The connected domain area Denoising Algorithm of bianry image can be described as:

1, according to 8 connected domains of pixel, bianry image is carried out zone marker, be divided into different connected regions；

2, the pixel in same area a kind of label, zones of different different labels；

3, according to different labels, it is added to corresponding array；

4, add up the number of each label, obtain the size of respective regions.If more than area threshold, this region is protected Staying, otherwise this region is set to background.

Fig. 3 (a) removes the result of block distortion for utilizing area.

After back eliminates small impurities region, a kind of noise region is also had to exist, that is, finger fingertip front end Region.Background, owing to finger fingertip is relatively thin, illumination is brighter, is considered as target during thresholding by column by this region, this part Region becomes maximum interference factor.The removal of this partial noise is determined by referring to position of cusp, and the row before finger tip are set to the back of the body Scape.

Human finger finger tip approximation circular arc, in theory during referring to that root scans by column to finger tip, closer to finger tip portion Point, the object pixel of these row and the fewest, crossed finger tip point, noise region each row white pixel point and increase suddenly, according to this Plant the regularity of distribution, it may be determined that go out and refer to position of cusp, thus by target and background separation.

First, the bianry image after removing edge block distortion carries out column direction projection, such as formula (3).In binary map In Xiang, represent background and target with 0 and 1, then the projection value of each column is this row object pixel number sum.

$L_i=\underset{j=0}{\overset{H-1}{\mathrm{\&Sigma;}}}p(i,j)---\left(3\right)$

Wherein, (i is j) that on bianry image, (i, j) value at place is 0 or 1, and H is the height of image to p.Therefore L_iRepresent i-th The number of object pixel in row.Fig. 3 (b) is the curve after vein image carries out column direction projection.

Find out that there are a good trough, corresponding Fingers cusp in drop shadow curve from Fig. 3 (b).According to statistics, different fingers are long Degree wave trough position is different, and majority is distributed between 80-150 row.In order to improve precision, the projection of 80-150 row in Fig. 3 (b) is taken The average of adjacent five row sums is as the value of these row.Then, from 150 row, right-to-left scans by column, when meeting trough condition Time, these row are the row at finger tip point place, and the row before finger tip point are set to background.Shown in effect such as Fig. 3 (c).

For reducing operand further, before image procossing below, first extract complete target area, subtract as far as possible Few background area.Extract and comprise the minimum rectangle of finger: the lateral length of rectangular area by the right side of the row of finger tip point and image certainly Fixed, the longitudinally wide of rectangular area is determined by the minima of finger top edge and the maximum of finger lower limb.Fig. 3 (d) shows Be extract finger areas image.

3, rotation correction

Because the collection of finger vena is contactless, not having the constraint of gim peg, finger is put naturally, same person Between different samples the most variant in horizontal offset, the anglec of rotation, these factors have impact on accuracy of identification.Therefore, first will Finger-image rotates to level, carries out ROI location the most again.

First having to determine the anglec of rotation of the vein image collected, the present invention is obtained by the slope of finger center line.? In bianry image, the marginal point of finger constitutes the profile of finger, and the point on finger center line is obtained by the midpoint of the upper and lower profile point of finger , these are put and simulates a near linear, then the slope of straight line is positive and negative relevant with direction of rotation, the size of slope and rotation Gyration is relevant.Fig. 4 (a) provides the some distribution on finger center line, and fitting a straight line.

Finger center line can use least square fitting to obtain.As a kind of optimization algorithm, method of least square is By making the error sum of squares trying to achieve between data and real data minimum, find optimum data.Intend corresponding to linear equation Close, it is simply that one group is met the data of linear equation, do difference with the linear equation set, during sum of squared errors function minimum, ask Obtain intercept a and the optimum of slope b.

If the expression formula of linear equation is:

y=a+bx (4)

A and b can be obtained by formula (5):

$a=\stackrel{\&OverBar;}{y}-b\stackrel{\&OverBar;}{x},b=\frac{\stackrel{\&OverBar;}{\mathrm{xy}}-\stackrel{\&OverBar;}{x}\stackrel{\&OverBar;}{y}}{\stackrel{\&OverBar;}{x^2}-{\stackrel{\&OverBar;}{x}}^2}---\left(5\right)$

Wherein:

$\stackrel{\&OverBar;}{x}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{x}_i,\stackrel{\&OverBar;}{y}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{y}_i$ (6)

$\stackrel{\&OverBar;}{x^2}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{x}_i^2,\stackrel{\&OverBar;}{\mathrm{xy}}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{x}_i{y}_i$

A is the intercept of straight line, and b is the slope of straight line, and the incline direction of finger and the anglec of rotation can be by the positive and negative of b and sizes Obtain.

Next according to the algorithm of image rotation, finger-image is adjusted to level.As shown in Fig. 4 (b).

4, anchor point determines

Containing two articulations digitorum manus on finger, finger being divided into three sections, we primarily determine that middle one section of extraction finger is sense Interest region.The rotation correction of finger areas, overcomes image rotation, bias effect, and next ROI orientation problem is discussed.Hands On finger, available information has finger length, width, but gathers due to finger every time and offset, and the finger length obtained is different, Therefore length information cannot function as the foundation of location.

The image finger tip contour approximation gathered is circular arc, and the diameter of circular arc therefore can be used to carry out ROI location.Concrete side Method is: first finds the effective circular arc point set on finger tip profile, then these points is justified matching, calculate matching diameter.

Effective finger tip profile point set is to be obtained by the method for rough estimate radius.Finger areas after rotation correction In, it is stipulated that finger center line left end is starting point, and the center of circle falls on finger center line, and the distance between starting point to the center of circle is radius.Circle The heart from left to right moves along finger center line from starting point, and radius is gradually increased, until radius beyond finger lower edges point away from Till linear distance, obtain the maximum estimated value of finger tip radius.As shown in Fig. 5 (a).Profile point in this radius is all effective Point.

Next these point sets are justified matching, by the way of calculating, determine the exact value of finger tip arc radius.Fig. 5 B () is the schematic diagram solving finger tip radius.Use Least-squares minimization algorithm.

If the point (X, Y) on circular arc meets equation of a circle:

R²=(X-A)²+(Y-B)² (7)

A, B, R can be obtained by formula:

$A=-\frac{a}{2}$

$B=-\frac{b}{2}---\left(8\right)$

$R=\frac{1}{2}\sqrt{a^2+b^2-4c}$

Wherein

$a=\frac{\mathrm{HD}-\mathrm{EG}}{\mathrm{CG}-D^2}$

$b=\frac{\mathrm{HC}-\mathrm{ED}}{D^2-\mathrm{GC}}---\left(9\right)$

$c=-\frac{\mathrm{\&Sigma;}(X_i^2+Y_i^2)+\mathrm{a\&Sigma;}{X}_i+\mathrm{b\&Sigma;}{Y}_i}{N}$

Formula (10) is shown in the calculating of C, D, E, G, H:

$C=(\mathrm{N\&Sigma;}{X}_i^2-\mathrm{\&Sigma;}{X}_i\mathrm{\&Sigma;}{X}_i)$

D=(N∑X_iY_i-∑X_i∑Y_i)

$E=\mathrm{N\&Sigma;}{X}_i^3+\mathrm{N\&Sigma;}{X}_i{Y}_i^2-\mathrm{\&Sigma;}(X_i^2+Y_i^2)\mathrm{\&Sigma;}{X}_i---\left(10\right)$

$G=(\mathrm{N\&Sigma;}{Y}_i^2-\mathrm{\&Sigma;}{Y}_i\mathrm{\&Sigma;}{Y}_i)$

$H=\mathrm{N\&Sigma;}{X}_i^2{Y}_i+\mathrm{N\&Sigma;}{Y}_i^3-\mathrm{\&Sigma;}(X_i^2+Y_i^2)\mathrm{\&Sigma;}{Y}_i$

R is that matching obtains radius, and finger tip circular diameter is anchor point.

5, region of interesting extraction

Finger tip profile radius of circle is tried to achieve, after rotation correction in finger areas, with finger center line left end for rising by matching Point, the cut-off rule on the left of interesting image regions vertical direction is: L1:x=2*R, and the cut-off rule L2 on right side is defined as according to L1: L2:x=2*R+d, in formula, d represents the ROI lateral length of extraction.According to statistics, the finger areas finger length after rotation correction exists About 200 pixels, finger tip a diameter of about 50, if referring to that root portion removes same length, then finger intermediate length is about 100, therefore d value is taken as 100.

In the contour images of finger areas, it is considered to the finger areas folded by cut-off rule L1, L2, the longitudinally width of ROI region The maximum spending the minima by finger contours top edge and finger contours lower limb determines, the rectangle that these four straight lines are enclosed is i.e. For ROI, this process schematic such as Fig. 6 (a) Suo Shi, the finger vena area-of-interest such as Fig. 6 (b) actually obtained is shown.

6, size normalization

The ROI width of said extracted is defined as 100, and different sample height are different, in order to study conveniently, ROI carried out height Degree normalization, makes the effective coverage that different sample is corresponding equivalently-sized.Size normalization is through the method for linear transformation, will Image not of uniform size becomes the image of prescribed level.

Size normalization relates to gray-level interpolation algorithm, this is because the point on target image differs establishes a capital corresponding original graph As upper integral point, it is frequently encountered by the situation that coordinate in length and breadth is all decimal, then the value of object pixel to be determined will be used slotting Value-based algorithm.In arest neighbors interpolation, bilinear interpolation [42] and curve interpolation, nearest neighbor method is the most simple and quick, for little The process of number is taked to round up and is found nearest integer；Bilinear interpolation is then to compose according to the distance to 4 consecutive points Upper corresponding weight value, calculates the value of this point, and error is less.It is considered as bilinearity method herein.

To a large amount of finger venous image experiment Analysis, the altitude range of area-of-interest is about 60, therefore, and will figure The height of picture is normalized to 60.It is the image of 100 × 60 that Fig. 7 is shown that finger vena ROI uniform sizes.

Claims (3)

1. a non-contact type finger vein sample area-of-interest exacting method, including Threshold segmentation by column, image denoising, rotation Transfer to another school just, anchor point determine, region of interesting extraction and size normalization, it is characterized in that: use varimax quiet to finger Arteries and veins image carries out Threshold segmentation by column, and carries out denoising by the method for connected domain area and searching finger tip, is partitioned into finger district Territory；Utilizing the upper and lower profile point of finger to simulate finger center line, the point on finger center line is obtained by the midpoint of the upper and lower profile point of finger , these points are simulated a near linear, using the slope of described finger center line as the foundation of rotation correction；Utilize finger tip The feature that contour approximation is round asks for the effective contour point set of circular arc, further with these effective contour point matching finger tips circle, To finger tip diameter of a circle, carry out area-of-interest location with this；The transverse width of area-of-interest is fixed width, longitudinally wide Determined by finger contours lower edges internal tangent；Finally the area-of-interest obtained is carried out size normalization.

Non-contact type finger vein sample area-of-interest exacting method the most according to claim 1, is characterized in that described Varimax finger vein image is used to carry out Threshold segmentation by column particularly as follows: light based on row each in image acquisition process According to being equally distributed, using every string of image as a unit, use varimax, calculate the segmentation threshold of these row Value, and carry out the global threshold segmentation of these row.

Non-contact type finger vein sample area-of-interest exacting method the most according to claim 1 and 2, is characterized in that institute State and carry out denoising particularly as follows: calculate each connected domain according to zone marker by the method for connected domain area and searching finger tip Area, retains the area in largest connected territory, eliminates little block distortion；For the noise of tip portion, according to each column in image Projection value on pixel vertical direction, determines the position of finger tip, eliminates the noise before finger tip.