CN102663393B - Method for extracting region of interest of finger vein image based on correction of rotation - Google Patents

Abstract

The invention aims at providing a method for extracting the region of interest of a finger vein image based on the correction of rotation. Firstly, dividing a finger region of a read-in finger vein image by employing a Kapur entropy threshold method; then, calculating a center of mass of the image and the center of mass is used as a reference for the correction of rotation, and determining a position of the region of interest based on the project values of pixels on each column of the image in a vertical direction and internal tangent lines of the upper and the lower edges of a finger outline; finally, normalizing samples of the image for obtaining a final processing result. Due to the problems that non-linear factors such as rotation, translation, etc. in the sampling of the finger vein image exert a significant impact on the image quality, and that the finger vein image positioning is difficult, the method brings up with a new solution which fully considers the characteristics of the non-contact sampling of the finger vein image and extracts the region of interest of the image based on the correction of rotation, and thus effectively improves influences of the image quality and enables the identification to be more reliable.

Description

Finger venous image area-of-interest exacting method based on rotation correction

Technical field

What the present invention relates to is a kind of vena characteristic extracting method of mode identification technology.

Background technology

Vein identification is as one authentication identifying method highly reliably, its recognition performance is closely related with the quality that vein target is extracted to a great extent, just can reach expection property effect because biometrics identification technology must be compared the feature of the same pattern of same area, this region is exactly region of interest ROI (Region of Interest).

The finger venous image area-of-interest obtaining for non-contact capture mode at present determines that the correlative study of aspect is less, to rely on to introduce positioning auxiliary device and palmmprint and the palm, dorsal vein interesting image regions are extracted to most of method, although document " hand-characteristic identification and feature level Study on Fusion " (Li Qiang. Beijing Jiaotong University's doctorate paper, 2006:22-24) proposed a kind of noncontact expansion palmmprint sample ROI extracting method, but the method still needs sample to have obvious finger uncut jade structural information.And finger elongated flat does not possess this characteristic, what the area-of-interest that makes said method cannot be used in reference to vein was determines.

Summary of the invention

The object of the present invention is to provide the finger venous image area-of-interest exacting method based on rotation correction of the impact that contributes to reduce accuracy on recognition system of the non-linear factor such as caused rotation, translation in sampling process and validity.

The object of the present invention is achieved like this:

The finger venous image area-of-interest exacting method that the present invention is based on rotation correction, is characterized in that:

(1) read in finger venous image, be partitioned into finger areas;

(2) ask for the barycenter of image, be rotated correction:

Barycenter C (the C of finger areas _x, C _y), its computing formula is as follows:

$C_x=\underset{i=0}{\overset{M}{\mathrm{\Σ}}}\underset{j=0}{\overset{N}{\mathrm{\Σ}}}{x}_i\×p(i,j)/\underset{i=0}{\overset{M}{\mathrm{\Σ}}}\underset{j=0}{\overset{N}{\mathrm{\Σ}}}p(i,j),$

$C_y=\underset{i=0}{\overset{M}{\mathrm{\Σ}}}\underset{j=0}{\overset{N}{\mathrm{\Σ}}}{y}_j\×p(i,j)/\underset{i=0}{\overset{M}{\mathrm{\Σ}}}\underset{j=0}{\overset{N}{\mathrm{\Σ}}}p(i,j),$

$p(i,j)=\left\{\begin{array}{c}1,(i,j)\∈I\\ 0,(i,j)\∉I\end{array}\right.,$

Wherein x _ithe horizontal ordinate of i pixel in presentation video, y _jthe ordinate of j element in presentation video, M presentation video wide, the height of N presentation video, belongs to the region of finger in I presentation video;

Obtaining after image centroid, finding the straight-line segment at last row place of finger contours image, and determining the middle point coordinate O of this line segment, tie point C and some O l in alignment _cO, calculated line l _cOwith horizontal direction line l _hangle be anglec of rotation θ, with this, image is rotated to correction;

The computing formula of θ is as follows:

θ＝tan ^-1(y _o-y _c)/(x _o-x _c)，

Wherein (x _c, y _c), (x _o, y _o) be respectively a C and the transverse and longitudinal coordinate figure of putting O, when θ > 0 is y ₀> y _ctime, image is turned clockwise; When θ < 0 is y ₀< y _ctime, image is rotated counterclockwise; When θ=0, i.e. y ₀=y _ctime, image is not rotated to operation;

If arbitrfary point A (x, y) is around rotation center C (C on image _x, C _y) be rotated counterclockwise θ degree, put the postrotational coordinate of A (x ', y ') and be:

x′＝(x-c _x)×cosθ+(y-c _y)×sinθ+c _x、

y′＝(y-c _y)×cosθ-(x-c _x)×sinθ+c _y，

If arbitrfary point A (x, y) is around rotation center C (C on image _x, C _y) the θ degree that turns clockwise, put the postrotational coordinate of A (x ', y ') and be:

x′＝(x-c _x)×cosθ-(y-c _y)×sinθ+c _x、

y′＝(y-c _y)×cosθ+(x-c _x)×sinθ+c _y；

(3) determine area-of-interest position:

Whole image-region is carried out to projection to vertical direction, i.e. the summation L of the gray-scale value of every row pixel in computed image _i:

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

The pixel that wherein p (i, j) lists for the capable j of i on image, the height that H is image,

In vertical direction projection, the moving window taking length as 15 is at L _ion carry out translation, in the pixel range of 0≤j < 180, find the region of mean value maximum, using the mid point p in this region as the articulation point of cutting apart, point the minimum horizontal ordinate of revolver profile, the cut-off rule in interesting image regions vertical direction left side is taken as: l ₁: x=p, the cut-off rule l on right side ₂be defined as according to l1: l ₂: x=p+d, in formula, d represents two vertical parallel lines l ₁, l ₂between distance, i.e. the transverse width of interesting image regions;

Utilize profile extraction algorithm to obtain the edge contour of the finger areas of image, obtain the image of Single pixel edge, then ask for respectively the internal tangent l at the upper and lower edge of finger contours ₃, l ₄, the straight line l of the straight line of these two horizontal directions and vertical direction ₁, l ₂the rectangular area of intersecting a sealing of formation is area-of-interest;

(4) image is carried out to sample normalization, obtain final result.

The present invention can also comprise:

1, the described finger areas of cutting apart operates to realize by Kapur entropy thresholding:

If f ₁, f ₂... f _nbe the pixel count of each gray level of a width finger venous image, wherein n is total number of greyscale levels, and destination probability distribution and background probability are distributed as with

Wherein: $p_i=\frac{f_i}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{f}_i},$ $P_t=\underset{i=1}{\overset{t}{\mathrm{\&Sigma;}}}{p}_i,$ $1-P_t=\underset{i=t+1}{\overset{n}{\mathrm{\&Sigma;}}}{p}_i,$ According to destination probability distribution and background probability definition posterior probability entropy be:

$\left\{\begin{array}{c}{H}_B\left(t\right)=-\underset{i=0}{\overset{t}{\mathrm{\&Sigma;}}}\frac{p_i}{P_t}{\log }_2\frac{p_i}{P_t}\\ H_w\left(t\right)=-\underset{i=t+1}{\overset{L-1}{\mathrm{\&Sigma;}}}\frac{p_i}{1-P_t}{\log }_2\frac{p_i}{1-P_t}\end{array}\right.,$

Entropy $H_B\left(t\right)+H_w\left(t\right)={\log }_2{P}_t(1-P_t)+\frac{H_t}{P_t}+\frac{H_{L-1}-H_t}{1-P_t}$ While reaching maximum, try to achieve optimal threshold t ^*, wherein $H_t=\underset{i=0}{\overset{t}{\mathrm{\&Sigma;}}}{p}_i{\log }_2{p}_i,$ $H_{L-1}=\underset{i=0}{\overset{L-1}{\mathrm{\&Sigma;}}}{p}_i{\log }_2{p}_i,$ According to optimal threshold t ^*image is carried out to thresholding, view data is divided into two parts: be greater than threshold value t ^*pixel portion and be less than threshold value t ^*pixel portion, establishing input picture is f (x, y), output image is f ' (x, y):

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

2, the value of the transverse width d of described interesting image regions gets 124.

Advantage of the present invention is: the present invention is that the non-linear factor such as rotation, translation existing in finger venous image sampling process proposes new solution thinking to picture quality impact problem large and finger venous image location difficulty, take into full account the feature of the contactless collection of finger venous image, the image collecting is carried out to the region of interesting extraction based on rotation correction, effectively improve the impact that collection image quality brings, made recognition result more reliable.

Brief description of the drawings

Fig. 1 is process flow diagram of the present invention;

Fig. 2 (a)-Fig. 2 (d) is that finger areas is extracted image; Wherein Fig. 2 (a) original image, the image after Fig. 2 (b) Kapur entropy Threshold segmentation, Fig. 2 (c) removes the image after burr, the finger areas image that Fig. 2 (d) extracts;

Fig. 3 key point schematic diagram;

Finger vein image after Fig. 4 rotation correction;

Fig. 5 (a)-Fig. 5 (d) is row grey scale pixel value curve; Wherein Fig. 5 (a) the position of the columns of getting in image, gray-scale value corresponding to Fig. 5 (b) a row pixel, gray-scale value corresponding to Fig. 5 (c) b row pixel, gray-scale value corresponding to Fig. 5 (d) c row pixel;

Fig. 6 is vertical direction projected image;

Fig. 7 is finger contours extraction effect;

Fig. 8 (a)-Fig. 8 (b) refers to vein ROI extracted region effect; The wherein structure in Fig. 8 (a) ROI region, the ROI region that Fig. 8 (b) extracts;

Image after Fig. 9 gray scale and size normalization.

Embodiment

For example the present invention is described in more detail below in conjunction with accompanying drawing:

In conjunction with Fig. 1～, 1 is partitioned into finger areas

Because the image collecting is subject to the impact of environment, the gray-scale value of its background pixel point is not 0 entirely, for fear of background, subsequent characteristics is extracted and the impact of identifying processing, needs the set of the pixel that obtains finger areas before subsequent treatment.The present invention operates to realize obtaining of finger areas by Kapur entropy thresholding.

If f2, f2 ... fn is the pixel count of each gray level of a width finger venous image, and wherein n is total number of greyscale levels, and destination probability distribution and background probability are distributed as with

Wherein: $p_i=\frac{f_i}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{f}_i},$ $P_t=\underset{i=1}{\overset{t}{\mathrm{\&Sigma;}}}{p}_i,$ $1-P_t=\underset{i=t+1}{\overset{n}{\mathrm{\&Sigma;}}}{p}_i.$ According to destination probability distribution and background probability definition posterior probability entropy be:

$\left\{\begin{array}{c}{H}_B\left(t\right)=-\underset{i=0}{\overset{t}{\mathrm{\&Sigma;}}}\frac{p_i}{P_t}{\log }_2\frac{p_i}{P_t}\\ H_w\left(t\right)=-\underset{i=t+1}{\overset{L-1}{\mathrm{\&Sigma;}}}\frac{p_i}{1-P_t}{\log }_2\frac{p_i}{1-P_t}\end{array}\right.---\left(1\right)$

In order to make target and background there is maximum fault information, must allow both posterior entropy sum for maximum, even entropy $H_B\left(t\right)+H_w\left(t\right)={\log }_2{P}_t(1-P_t)+\frac{H_t}{P_t}+\frac{H_{L-1}-H_t}{1-P_t}$ While reaching maximum, can try to achieve optimal threshold t ^*, wherein $H_t=\underset{i=0}{\overset{t}{\mathrm{\&Sigma;}}}{p}_i{\log }_2{p}_i,$ $H_{L-1}=\underset{i=0}{\overset{L-1}{\mathrm{\&Sigma;}}}{p}_i{\log }_2{p}_i.$ According to optimal threshold t ^*image is carried out to thresholding, view data is divided into two parts: be greater than threshold value t ^*pixel portion and be less than threshold value t ^*pixel portion.If input picture is f (x, y), output image is f ' (x, y):

$f^{\&prime;}(x,y)=\left\{\begin{array}{c}1,f(x,y)\&GreaterEqual;t^{*}\\ 0,f(x,y)<t^{*}\end{array}\right.---\left(2\right)$

Experiment original image is as shown in Fig. 2 (a), and the result after thresholding is as shown in Fig. 2 (b), and the white portion in image is the finger areas extracting.Owing to having burr in the image after Threshold segmentation, adopt the operation of opening in mathematical morphology to process as shown in Fig. 2 (c) it herein, the finger areas image finally extracting is as shown in Fig. 2 (d), and follow-up processing herein all adopts the image extracting after finger areas to carry out.

2 rotation corrections

While referring to vein image acquisition, because gathered person points putting position and direction is some difference, make the vein image that different time obtains from same finger have rotation and translation phenomenon in various degree, and finger is not such as the feature that refers to the auxiliary ROI extracted region such as concave, convex point that uncut jade or finger-joint are bent to form, therefore, the present invention proposes first finger vein image before ROI extracted region and is rotated correction, on the basis of rotation correction, extract the ROI region of image, can reduce the difficulty of successive image coupling, increase the robustness of system.

The barycenter of finger areas is the performance index that every width image all exists, and the calculating of barycenter is of overall importance, and its antijamming capability is stronger, and therefore the present invention is rotated correction by the barycenter in display foreground region.Extract after finger areas in the method that adopts said extracted finger areas, calculating target image is the barycenter C (C of finger areas _x, C _y), its computing formula is as follows:

$C_x=\underset{i=0}{\overset{M}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{N}{\mathrm{\&Sigma;}}}{x}_i\&times;p(i,j)/\underset{i=0}{\overset{M}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{N}{\mathrm{\&Sigma;}}}p(i,j)---\left(3\right)$

$C_y=\underset{i=0}{\overset{M}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{N}{\mathrm{\&Sigma;}}}{y}_j\&times;p(i,j)/\underset{i=0}{\overset{M}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{N}{\mathrm{\&Sigma;}}}p(i,j)---\left(4\right)$

$p(i,j)=\left\{\begin{array}{c}1,(i,j)\&Element;I\\ 0,(i,j)\&NotElement;I\end{array}\right.---\left(5\right)$

Wherein, x _ithe horizontal ordinate of i pixel in presentation video, y _jthe ordinate of j element in presentation video, M presentation video wide, the height of N presentation video, belongs to the region of finger in I presentation video.

Obtaining after image centroid, we find the straight-line segment at last row place of finger contours image, and determine the middle point coordinate O of this line segment, tie point C and some O l in alignment _cO, calculated line l _cOwith horizontal direction line l _hangle be anglec of rotation θ, (as shown in Figure 3) is rotated correction with this to image.The computing formula of θ is as follows:

θ＝tan ^-1(y _o-y _c)/(x _o-x _c)(6)

Wherein (x _c, y _c), (x _o, y _o) be respectively a C and some O transverse and longitudinal coordinate figure.As θ > 0, i.e. y ₀> y _ctime, image is turned clockwise; As θ < 0, i.e. y ₀< y _ctime, image is rotated counterclockwise; When θ=0, i.e. y ₀=y _ctime, image is not rotated to operation.

If arbitrfary point A (x, y) is around rotation center C (C on image _x, C _y) be rotated counterclockwise θ degree, put the postrotational coordinate of A (x ', y ') and be:

x′＝(x-c _x)×cosθ+(y-c _y)×sinθ+c _x (7)

y′＝(y-c _y)×cosθ-(x-c _x)×sinθ+c _y (8)

If arbitrfary point A (x, y) is around rotation center C (C on image _x, C _y) the θ degree that turns clockwise, put the postrotational coordinate of A (x ', y ') and be:

x′＝(x-c _x)×cosθ-(y-c _y)×sinθ+c _x (9)

y′＝(y-c _y)×cosθ+(x-c _x)×sinθ+c _y (10)

Image after rotation correction, as shown in Figure 4.

3 determine area-of-interest position

This be in that above-mentioned finger areas is extracted and the basis of rotation correction on complete determining of ROI region to referring to vein image.Because articulations digitorum manus position has cartilaginous tissue, irradiating and obtaining under the mode of image based near infrared, articulations digitorum manus has stronger penetration capacity with respect to other position of finger, therefore, in whole finger vein image, finger-joint position brightness ratio is larger, and the pixel value of this parts of images is high compared with other parts.

We extract the pixel at a row articulations digitorum manus position from image array, extract again a row pixel from other position, the gray-scale value of this two row pixel is plotted to curve map as shown in Fig. 5 (b), Fig. 5 (c) and Fig. 5 (d), in figure, horizontal ordinate is the line number at institute's capture element place, the gray-scale value that ordinate is this pixel.From figure, can draw, the gray-scale value of that row pixel of joint part wants high with respect to non-joint part.Therefore, the present invention utilizes joint position that image is positioned and cut apart.

First, whole image-region is carried out to projection to vertical direction, i.e. the summation L of the gray-scale value of every row pixel in computed image _i:

$L_i=\underset{j=0}{\overset{H-1}{\mathrm{\&Sigma;}}}p(i,j)---\left(11\right)$

Wherein, the pixel that p (i, j) lists for the capable j of i on image, the height that H is image.

Curve after image projection as shown in Figure 6, at first articulations digitorum manus place, curve has a good peak value, be that 60 in correspondence image is listed as to 90 row, and the peak value of second articulations digitorum manus is obvious not as first, this is that in muscle, moisture is higher because the thickness of second articulations digitorum manus position muscle wants large with respect to first articulation point, moisture has certain absorption to infrared ray, thereby causes the peak value on curve not obvious.Therefore, the present invention chooses the location cut-point of first articulation point as area-of-interest.

In vertical direction projection, the moving window taking length as 15 is at L _ion carry out translation, in the pixel range of 0≤j < 180, find the region of mean value maximum, using the mid point p in this region as the articulation point of cutting apart, point the minimum horizontal ordinate of revolver profile, the cut-off rule in interesting image regions vertical direction left side is taken as: l ₁: x=p, the cut-off rule l on right side ₂according to l ₁be defined as: l ₂: x=p+d, in formula, d represents two vertical parallel lines l ₁, l ₂between distance, i.e. the transverse width of interesting image regions.In theory, d value is chosen larger, and the area of the area-of-interest of image is larger, and the quantity of information comprising in this sampled images is just more, is conducive to coupling, the identifying operation of successive image.But, we comprehensively analyze discovery to a large amount of finger vein images, different acquisition person's finger length difference, even same picker, in the time of image acquisition, the length difference of pointing in the image that the position difference of putting due to finger can cause collecting, makes the position of the first articulations digitorum manus in image have certain difference, if it is excessive that d value is chosen, will exist the value of p+d to exceed the possibility of image range.Therefore, in order to take into account area-of-interest area as far as possible greatly and cut-off rule l ₂do not exceed image range, d value is taken as to 124.

In order to determine the cut-off rule of image level direction, first utilize profile extraction algorithm to obtain the edge contour of the finger areas of image, obtain the image of Single pixel edge as shown in Figure 7.

Then ask for respectively the internal tangent l at the upper and lower edge of finger contours ₃, l ₄, the straight line l of the straight line of these two horizontal directions and vertical direction ₁, l ₂intersect the rectangular area that forms a sealing as shown in Fig. 8 (a), adopt the mode of internal tangent to avoid outer tangential way that background area is sneaked in area-of-interest, the ROI region of the finger vena extracting is as shown in Fig. 8 (b).

4 sample normalization

In sum, obtain sample image owing to adopting contactless mode, and this mode does not limit pointing plane and the distance of camera and the swing of finger level, there is very large uncertainty in the putting position of therefore pointing when collecting sample, thereby cause the finger vein image collecting not only to have the difference in direction, also there is the difference in gray scale and size, in order to reduce the impact of these factors, improve the performance of vein recognition system, need to carry out gray scale and yardstick normalization operation to vein image.

4.1 gray scale normalizations:

Owing to adopting the mode of near-infrared light source to refer to the collection of vein sample image, make the picture contrast that collects lower, gradation of image distributes relatively concentrated; In addition, pointed the impact of the factor such as distance and visible ray of plane and camera, also can cause vein image intensity profile situation to have certain difference, therefore need image to carry out gray scale normalization.The object of gray scale normalization is, by the gray-scale value computing of pixel, image is carried out to gray scale stretching, and the gray-scale value scope of pixel is dispersed in whole gray levels, the image strengthening to obtain light and shade contrast.The normalization formula that the present invention adopts is:

F (x, y)=((g (x, y)-g _min(x, y)) × 255)/(g _max(x, y)-g _min(x, y)) in (12) formula, g (x, y), f (x, y) represent respectively the gray-scale value of sample image pixel before and after gray scale normalization, g _max(x, y), g _min(x, y) represents respectively maximum gradation value and the minimum gradation value of the front sample image pixel of normalization.

4.2 size normalization:

In the image that causes collecting due to factors such as finger plane and the distance of camera and different acquisition person's finger difference in size there is certain difference in finger areas size, therefore need image to carry out size normalization operation, make the effective coverage consistent size of image.

The normalized essence of size is that image is carried out to linear transformation, makes the effective coverage that different samples are corresponding have identical size by conversion.Size normalization conversion is the conventional conversion during image is processed, and does not do detailed introduction at this, but its key is to the determining of picture size after normalization, if size Selection is too large, can increase the pixel number of image after conversion, increases the calculated amount of subsequent algorithm; And size Selection is too little, can cause again effective image resolution is the loss of quantity of information not, is unfavorable for equally the processing that recognition system is follow-up.

Image after above-mentioned region of interesting extraction is rectangle, and the length of this rectangular area is defined as 124, only considers the value of its height herein.To a large amount of finger vein image experiment Analysis, the altitude range of area-of-interest is between 57-65, and wherein great majority concentrate near 62, therefore, the height of image is normalized to 64, Figure 9 shows that the design sketch after gradation of image normalization and size normalization.

5 interpretations

In order to verify the validity of this paper method, select laboratory set up finger vein image storehouse in image test.The finger vein image that this storehouse comprises 300 people, wherein everyone gathers fingerprint image 5 width, totally 1500 width, image size is.

For abundant checking this paper method performance, adopt matching process based on details respectively all images in original image storehouse and the ROI area image storehouse that extracts thereof to be carried out to certification in 1: 1 is tested and 1: n identifies experiment.When experiment, from everyone 5 width vein images, an optional width (totally 300 width images) is as composition of sample checking to be identified storehouse, and all the other 4 width (totally 300 × 4=1200 width image) form template base, and experimental result is as shown in table 1 and table 2.

Table 1

Table 2

Can draw from above-mentioned experimental result, adopt ROI method for extracting region based on rotation correction not only to improve the discrimination of system, and reduced reject rate and misclassification rate, system performance be improved significantly, thereby proved the validity of the inventive method.

Claims (3)

1. the finger venous image area-of-interest exacting method based on rotation correction, is characterized in that:

(1) read in finger venous image, be partitioned into finger areas;

(2) ask for the barycenter of image, be rotated correction:

Barycenter C (the C of finger areas _x, C _y), its computing formula is as follows:

$C_x=\underset{i=0}{\overset{M}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{N}{\mathrm{\&Sigma;}}}{x}_i\&times;p(i,j)/\underset{i=0}{\overset{M}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{N}{\mathrm{\&Sigma;}}}p(i,j),$

$C_y=\underset{i=0}{\overset{M}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{N}{\mathrm{\&Sigma;}}}{y}_j\&times;p(i,j)/\underset{i=0}{\overset{M}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{N}{\mathrm{\&Sigma;}}}p(i,j),$

$p(i,j)=\left\{\begin{array}{c}1,(i,j)\&Element;I\\ 0,(i,j)\&NotElement;I\end{array}\right.,$

Wherein x _ithe horizontal ordinate of i pixel in presentation video, y _jthe ordinate of j element in presentation video, M presentation video wide, the height of N presentation video, belongs to the region of finger in I presentation video;

Obtaining after image centroid, finding the straight-line segment at last row place of finger contours image, and determining the middle point coordinate O of this line segment, tie point C and some O l in alignment _cO, calculated line l _cOwith horizontal direction line l _hangle be anglec of rotation θ, with this, image is rotated to correction;

The computing formula of θ is as follows:

θ＝tan ^-1(y _o-y _c)/(x _o-x _c)，

Wherein (x _c, y _c), (x _o, y _o) be respectively a C and the transverse and longitudinal coordinate figure of putting O, when θ >0 is y ₀>y _ctime, image is turned clockwise; When θ <0 is y ₀<y _ctime, image is rotated counterclockwise; When θ=0, i.e. y ₀=y _ctime, image is not rotated to operation;

If arbitrfary point A (x, y) is around rotation center C (C on image _x, C _y) be rotated counterclockwise θ degree, put the postrotational coordinate of A (x ', y ') and be:

x′＝(x-c _x)×cosθ+(y-c _y)×sinθ+c _x、

y′＝(y-c _y)×cosθ-(x-c _x)×sinθ+c _y，

If arbitrfary point A (x, y) is around rotation center C (C on image _x, C _y) the θ degree that turns clockwise, put the postrotational coordinate of A (x ', y ') and be:

x′＝(x-c _x)×cosθ-(y-c _y)×sinθ+c _x、

y′＝(y-c _y)×cosθ+(x-c _x)×sinθ+c _y；

(3) determine area-of-interest position:

Whole image-region is carried out to projection to vertical direction, i.e. the summation L of the gray-scale value of every row pixel in computed image _i:

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

The gray-scale value of the pixel that wherein p (i, j) lists for the capable j of i on image, the height that H is image,

In vertical direction projection, the moving window taking length as 15 is at L _ion carry out translation, in the pixel range of 0≤j<180, find the region of mean value maximum, using the mid point p in this region as the articulation point of cutting apart, point the minimum horizontal ordinate of revolver profile, the cut-off rule in interesting image regions vertical direction left side is taken as: l ₁: x=p, the cut-off rule l on right side ₂according to l ₁be defined as: l ₂: x=p+d, in formula, d represents two vertical parallel lines l ₁, l ₂between distance, i.e. the transverse width of interesting image regions;

Utilize profile extraction algorithm to obtain the edge contour of the finger areas of image, obtain the image of Single pixel edge, then ask for respectively the internal tangent l at the upper and lower edge of finger contours ₃, l ₄, the straight line l of the straight line of these two horizontal directions and vertical direction ₁, l ₂the rectangular area of intersecting a sealing of formation is area-of-interest;

(4) image is carried out to sample normalization, obtain final result.

2. the finger venous image area-of-interest exacting method based on rotation correction according to claim 1, is characterized in that: the described finger areas of cutting apart operates to realize by Kapur entropy thresholding:

For tonal range 0,1 ... the finger venous image of L-1}, establishes f ₀, f ₁... f _l-1for the pixel count of each gray level, its destination probability distributes and background probability is distributed as with $\frac{p_{t+1}}{1-P_t},\frac{p_{t+2}}{1-P_t},...,\frac{p_{L-1}}{1-P_t}$ Wherein: $p_i=\frac{f_i}{\underset{i=0}{\overset{L-1}{\mathrm{\&Sigma;}}}{f}_i},P_t=\underset{i=0}{\overset{t}{\mathrm{\&Sigma;}}}{p}_i,1-P_t=\underset{i=t+1}{\overset{L-1}{\mathrm{\&Sigma;}}}{p}_i,$ According to destination probability distribution and background probability definition posterior probability entropy be:

$\left\{\begin{array}{c}{H}_B\left(t\right)=\underset{i=0}{\overset{t}{\mathrm{\&Sigma;}}}\frac{p_i}{P_t}{\log }_2\frac{p_i}{P_t}\\ H_W\left(t\right)=-\underset{i=t+1}{\overset{L-1}{\mathrm{\&Sigma;}}}\frac{p_i}{1-P_t}{\log }_2\frac{p_i}{1-P_t}\end{array}\right.,$

Entropy $H_B\left(t\right)+H_w\left(t\right)={\log }_2{P}_t(1-P_t)+\frac{H_t}{P_t}+\frac{H_{L-1}-H_t}{1-P_t}$ While reaching maximum, try to achieve optimal threshold t ^*, wherein $H_t=\underset{i=0}{\overset{t}{\mathrm{\&Sigma;}}}{p}_i{\log }_2{p}_i,H_{L-1}=\underset{i=0}{\overset{L-1}{\mathrm{\&Sigma;}}}{p}_i{\log }_2{p}_i,$ According to optimal threshold t ^*image is carried out to thresholding, view data is divided into two parts: be greater than threshold value t ^*pixel portion and be less than threshold value t ^*pixel portion, establishing input picture is f (x, y), output image is f'(x, y):

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

3. the finger venous image area-of-interest exacting method based on rotation correction according to claim 1 and 2, is characterized in that: the value of the transverse width d of described interesting image regions gets 124.