CN103903001A - Finger vein network accurate extracting method - Google Patents

Finger vein network accurate extracting method Download PDF

Info

Publication number
CN103903001A
CN103903001A CN201410102276.0A CN201410102276A CN103903001A CN 103903001 A CN103903001 A CN 103903001A CN 201410102276 A CN201410102276 A CN 201410102276A CN 103903001 A CN103903001 A CN 103903001A
Authority
CN
China
Prior art keywords
image
algorithm
finger
obtains
finger vena
Prior art date
Application number
CN201410102276.0A
Other languages
Chinese (zh)
Inventor
杨金锋
师一华
李承尚
Original Assignee
中国民航大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 中国民航大学 filed Critical 中国民航大学
Priority to CN201410102276.0A priority Critical patent/CN103903001A/en
Publication of CN103903001A publication Critical patent/CN103903001A/en

Links

Abstract

A finger vein network accurate extracting method sequentially includes the steps of firstly, enhancing an original finger vein image; secondly, processing an enhanced image obtained in the first step through an automatic marking algorithm to obtain a trimap image; thirdly, extracting a clear finger vein network through an image matting algorithm with the trimap image obtained in the second step serving as a constraint. By means of the finger vein network accurate extracting method, the finger vein network can be accurately presented, and thin veins can be found. Thus, the finger vein network accurate extracting method has great significance in presenting the finger vein network.

Description

The accurate extracting method of a kind of finger vena network

Technical field

The invention belongs to Computer Recognition Technology field, particularly relate to the accurate extracting method of a kind of finger vena network.

Background technology

Because finger vena is positioned at skin bottom, therefore need to adopt the mode of transmission to carry out imaging.But, in imaging process, because there are the effects such as absorption, reflection, scattering and refraction in the biological tissues such as bone, muscle, vascular tissue's liquid and skin near infrared light, thereby after causing near infrared light to see through finger, produce power decays, it is the low-quality image that a kind of serious degradation is fuzzy that result causes the finger venous image collecting, and can not clearly reflect vein network.Therefore,, in order to extract exactly architectural feature, need to carry out a series of processing to obtain clear reliable vein network to original vein image.

At present less for the research of low-quality vein image extracting method, take to cut apart or the mode of feature extraction more, mainly contain traditional thresholding method, maximum principal curvatures method, repeat linear tracking etc.Obviously, the traditional images dividing method including the linear track algorithm of repetition all cannot accurately represent finger vena network, nor can excavate thinner vein.

Summary of the invention

In order to address the above problem, the object of the present invention is to provide the accurate extracting method of a kind of finger vena network.

In order to achieve the above object, the accurate extracting method of finger vena network provided by the invention comprises the following step of carrying out in order:

1) original finger venous image is strengthened;

2) enhancing imagery exploitation automatic mark algorithm step 1) being obtained is processed, and obtains trimap image;

3) take step 2) the trimap image that obtains is constraint, adopts image to take algorithm and extracts finger vena network clearly.

The method in described step 1), original finger venous image being strengthened is to utilize mist elimination Processing Algorithm to carry out scattered noise removal, obtain original mist elimination image R (x, y), then through multiple dimensioned even Gabor filtering strengthen the operator image E (x, y) that is enhanced.

Described step 2) in utilize the enhancing image that automatic mark algorithm obtains step 1) to process, the method that obtains trimap image is, the enhancing image E (x obtaining with step 1), y) be basis, utilize K-means to carry out clustering to strengthening image, successively generate the binary map of interim prospect and background by classification mark, recycle dynamic bianry image, generate the signature ζ of non-cross overlapping foreground pixel set through two-value operation operator fthe signature ζ of (x, y) and background pixel set b(x, y), afterwards with the signature ζ of foreground pixel set fthe signature ζ of (x, y) and background pixel set b(x, y) generates scribble district for constraining on original image, obtains retraining image C (x, y).

In described step 3), utilize image to take algorithm to step 2) method extracted of the trimap image that obtains is first with constraint image C (x, y) estimate to obtain shade α, and then extract foreground picture as constraint by Optimized Iterative method take shade α, obtain thus finger vena network clearly.

The accurate extracting method of finger vena network provided by the invention can accurately represent finger vena network, and can excavate thinner vein.Therefore, this method has great importance for the expression of finger vena network.

Accompanying drawing explanation

Fig. 1 is the workflow diagram of the accurate extracting method of finger vena network provided by the invention.

Fig. 2 is the tangent schematic diagram of spatial frequency domain profile.

Fig. 3 is the spatial domain response of one group of even symmetry Gabor wave filter.

Embodiment

Below in conjunction with the drawings and specific embodiments, the accurate extracting method of finger vena network provided by the invention is elaborated.

As shown in Figure 1, the accurate extracting method of finger vena network provided by the invention comprises the following step of carrying out in order:

1) original finger venous image is strengthened;

Due to the impact of the external factors such as absorption, reflection, scattering, make the finger venous image contrast that collects lower, first venous structures impalpable, therefore, need to carry out figure image intensifying; The object of figure image intensifying mainly contains two aspects: one, improve the visual effect of image, improve the sharpness of image; Two, give prominence to selectively the interested information of people, suppress garbage, thereby improve the use value of image, and image is converted to the form of a kind of people of being more suitable for or machine processing.The present invention uses mist elimination algorithm to realize first object, uses one group of special even Gabor wave filter of inventor's design to realize second object.

Medical research shows, bone, muscle, vascular tissue's liquid and the skin histology of finger have the effects such as absorption, reflection, refraction, scattering near infrared light.This is just similar to, and sharply reduces this phenomenon in the greasy weather because the scattering process of atmospheric particles makes imaging scene visibility.Therefore the phenomenon that, causes finger venous image to be degenerated due to the strong scattering characteristic of biological tissue is inevitable.So, certainly, we just can adopt the method for removing scattered noise to rebuild image, improve picture quality.

At present, in multinomial application, the image mist elimination technology based on Koschmieder law is proved to be a kind of method that can effectively remove scattered noise.Suppose that I (x, y) is for we take the image obtaining, R (x, y) be original mist elimination image, ρ (λ) represents the extinction coefficient of atmosphere, d (x, y) be scene depth, can obtain according to Koschmieder law:

I(x,y)=R(x,y)e -ρ(λ)d(x,y)+I s(1-e -ρ(λ)d(x,y))????(1)

Wherein, λ is light wavelength, I sfor the intensity of illumination of surrounding environment.Airborne particulate and biological tissue can regard two kinds of scattering mediums different wavelength to different extinction coefficients as, and therefore, Koschmieder law can apply to the recovery of finger venous image.

But in solving practical problems, we are difficult to obtain extinction coefficient ρ (λ) and the scene depth d (x, y) of atmosphere accurately, this just makes above formula become an ill problem.Here, a kind of method direct estimation airlight intensity υ (x, y) based on wave filter that we adopt Jean-Philippe and Nicolas to propose in 2009,

υ(x,y)=I s(1-e -ρ(λ)d(x,y))????(2)

And no longer estimate extinction coefficient ρ (λ), the scene depth d (x, y) of atmosphere and the intensity of illumination I of surrounding environment s.Suppose the intensity of illumination I of surrounding environment sbe a constant, we solve about the equation of original mist elimination image R (x, y) (1) and just can obtain:

R(x,y)=I s(I(x,y)-υ(x,y))/(I s-υ(x,y))????(3)

This method can be restored fast to a secondary finger venous image, effectively strengthens the sharpness of image.

The removal of scattered noise can significantly improve picture quality.But, due to the complicacy of biological tissue, fuzzy still very serious between venosomes and non-vein region.Therefore, for the sharpness of further enhancing venosomes, improve the contrast in venosomes and non-vein region, the enhancing of blood vessel remains a challenging problem.Vein shows with carinate form under near infrared optical transmission, and the algorithm for image enhancement based on ridge can be explored venous information better so.Because Gabor wave filter is all adjustable in yardstick and direction, so, we for this Design of Problems one group of special even Gabor wave filter.

A two-dimentional Gabor wave filter is a function by Gaussian function and complex plane wave component:

G ( x , y ) = γ 2 π σ 2 exp { - 1 2 ( x θ 2 + γ 2 y θ 2 σ 2 ) } exp ( j ^ 2 π f 0 x θ )

x θ y θ = cos θ sin θ - sin θ cos θ x y - - - ( 4 )

Wherein, θ is the direction of Gabor wave filter, f 0for the centre frequency of wave filter, σ and γ represent respectively the aspect ratio of standard deviation and oval Gaussian envelope line, x θ, y θbe respectively coordinate x, the coordinate after y rotation.Utilize Euler's formula, we can be decomposed into function G (x, y) real part and imaginary part two parts.Real part is referred to as even symmetry Gabor wave filter, is suitable for ridge and detects; Imaginary part is referred to as odd symmetry Gabor wave filter, is suitable for rim detection.

Vein form with ridge in image shows, and therefore, we use one group of even symmetry Gabor wave filter to detect vein blood vessel, are deducted DC response and are:

G mk e ( x , y ) = γ 2 π σ m 2 exp { - 1 2 ( x θ k 2 + γ 2 y θ k 2 σ m 2 ) × ( cos ( 2 π f m x θ k ) - exp ( - v 2 2 ) ) - - - ( 5 )

Wherein, m is yardstick index, and k is direction index, and v is the DC response factor.Fig. 2 is the tangent schematic diagram of spatial frequency domain profile, and as shown in Figure 2, in spatial frequency domain, any two half amplitude point profiles are tangent along radial frequency or direction, and its parameter need to meet the following conditions:

Wherein,

K a = ( 2 Δω - 1 ) / ( 2 Δω + 1 ) ζ = ( 1 + K a ) / ( 1 - K a ) - - - ( 7 )

A m4) be the minor axis of Gabor frequency response half amplitude point profile under m yardstick, △ ω (∈ [0.5,2.5]) is frequency bandwidth, for half amplitude point direction bandwidth.For a specific frequency bandwidth, K afix, and, meet DC response subtraction.Right carry out Fourier transform, we can obtain:

a m = γ 21 n 2 σ m π - - - ( 8 )

From formula (6), we can obtain:

σ m f m = 1 K a π ln 2 2 - - - ( 9 )

According to Fig. 2, from formula (8) and formula (9), we just can derive:

Make N one determine that the number of irredundant profile on yardstick meets simultaneously by formula (10), the aspect ratio γ of oval Gaussian envelope line is approximately:

γ ≈ sin ( π 2 N ) / K a - - - ( 11 )

Therefore, as long as determined three parameter △ ω, σ 1the number of (out to out) and N(direction), we just can design one group of break-even even symmetry Gabor wave filter successively.Fig. 3 has shown one group of spatial domain response with the even symmetry Gabor wave filter of three centre frequencies and eight directions.

Based on even symmetry Gabor bank of filters, we carry out two-dimensional convolution to a sub-picture just can obtain different filtering images.Strengthen algorithm in order more clearly to set forth follow-up vein, have central peak Two-Dimensional Gabor Wavelets we be referred to as " main lobe ", the minor beam on side is referred to as " secondary lobe ".

Order for the negate figure of restored image R (x, y), for the filtering result in even symmetry Gabor filter transform territory, have:

U m k ( x , y ) = G mk e ( x , y ) ⊗ R ^ ( x , y ) - - - ( 12 )

Why we use the negate figure of restored image R (x, y) because it can keep the consistance between brightness and the even symmetry Gabor small echo main lobe of ridge in carrying out convolution.Making respectively M and N is scale parameter and direction number, and we just can obtain M × N width and have the filtering image of spatial locality and directional selectivity so.

Suppose p (x i, y i) be on a pixel, for p (x i, y i) N the point set converting on m yardstick.Work as θ k, σ mwhile matching partly the direction of vein and width simultaneously, obtain maximal value.Therefore, if make Z m(x i, y i) be the optimal response of a sub-picture on m yardstick, just can obtain:

Z m ( x i , y i ) = arg max U m k ( x i y i ) ∈ S m i i ∈ Λ ( U m k ( x i , y i ) ) - - - ( 13 )

Wherein, Λ is the index set of finger venous image pixel, S m i = { U m 0 ( x i , y i ) , L , U m k ( x i , y i ) , L , U m N - 1 ( x i , y i ) } .

Certainly, Z m(x, y) can comprise the information of vein crestal line on m yardstick completely, and still, it has also comprised the false venous information much being strengthened by secondary lobe simultaneously.And, the Z of single yardstick m(x, y) cannot describe the variation of vein diameter.Therefore, consider the variation of vein and the elimination of false vein, we propose a kind of multiple dimensioned multiplication criterion (MSMR), are defined as follows:

E ( x , y ) = Π m = 1 M Z m ( x , y ) - - - ( 14 )

E (x, y) is final enhancing image.

2) enhancing imagery exploitation automatic mark algorithm step 1) being obtained is processed, and obtains trimap image;

The feature in finger vena with the property distinguished and robustness is mainly included in venous structures, so topology-preserving is very important for our vein image is cut apart to greatest extent.But traditional finger vena diakoptic algorithm can not extract complete vein network.Although the MSMR that upper joint proposes can promote the contrast between venosomes and non-vein region well,, only use in actual applications MSMR, we are still difficult to obtain than more complete vein network.Trace it to its cause, mainly contain 2 points: the brightness of (1), vein crestal line changes greatly; (2), have that vein crestal line that larger brightness changes normally intersects.Therefore we need to utilize automatic mark algorithm to do further processing to strengthening image E (x, y).Automatic mark algorithm comprises three steps: K-means clustering algorithm, generation Framework Arithmetic and generation scribble district algorithm.

One, K-means cluster

Suppose that F (x, y) and B (x, y) represent respectively foreground picture and Background, a width finger vena palinspastic map just can be expressed as:

R ^ ( x , y ) = F ( x , y ) α ( x , y ) + B ( x , y ) ( 1 - α ( x , y ) ) - - - ( 15 )

Wherein, the prospect opacity that α (x, y) is pixel.For simplicity, and α (x, y) is expressed as in follow-up narration and α.In fact, F, B and α are unknown.But, if we have known the rough division of front and back background and zone of ignorance, just can estimate shade α according to formula (15).Here, we adopt the method for K-means automatically to generate trimap image (one is divided roughly).

Making K is pixel cluster numbers, D is the Euclidean distance of any two adjacent cluster centres, strengthen finger venous image E (x at a width, y) we just can obtain K logic bianry image (LBIs) above to use k-means clustering algorithm, and every piece image represents a corresponding pixel class.Suppose h j(x, y) represents j LBI, h 1(x, y), corresponding to first pixel class with center of maximum value, we define:

H l ( x , y ) = Σ j = 1 l h j ( x , y ) H ‾ l ( x , y ) = 1 - H l ( x ) - - - ( 16 )

Wherein, l=1, L, K-1, H l(x, y) is also a LBI, H lpixel representative in (x, y) is by front l the l that pixel class forms interim prospect, so a H lthe reverse image of (x, y) represent l interim background.Here must meet K × D≤255, to guarantee that classification is rational.

Two, generate skeleton

To bianry image H l(x, y) and we just can obtain the skeleton diagram of prospect and background to use morphological operation, are respectively ω l(x, y) and we define:

Wherein, BW () is binaryzation operational character.Obviously, ζ f(x, y) and ζ b(x, y) represents respectively the set of all K-1 interim prospects and interim background.

Can find out from formula (16), due to l<K-1, so in necessarily include a part of H l(x, y), that is to say foreground image and background image some juxtaposition that is bound to, and this distributes for appropriate pixel tag is unallowed.For addressing this problem, we generate following dynamically two-value background:

&kappa; T ( x , y ) = &Sigma; t = 1 T h K - t ( x , y ) - - - ( 18 )

Wherein, T representative forms the variable number of the LBI of a width finger venous image background, h k(x, y) is corresponding to last LBI.From κ tthe skeleton obtaining in (x, y) can effectively be removed lap.Therefore, we define:

z T ( x , y ) = &zeta; b ( x , y ) &CircleTimes; &kappa; T ( x , y ) - - - ( 19 )

Wherein, represent AND-operation.Make Q tfor corresponding respectively to ζ f(x, y) and z tthe non-null intersections set of two pixel clusters of (x, y), dynamically determine suitable T by following formula:

T = arg min T < K ( Q T ) - - - ( 20 )

Thus, ζ bthe part that is present in foreground area in (x, y) just can successfully be eliminated.

Three, generate scribble district

Based on formula (17) and formula (19), using the reverse image of finger venous image of removing scattered noise as with reference to figure, we can build piece image in the following way:

Suppose to represent respectively corresponding to two pixel class of C (x, y)=1 and C (x, y)=0 middle part prospect and part background will automatically generate scribble region so in a width finger venous image.Scratch figure for image, C (x, y) is regarded as a constraints graph picture or trimap image (i.e. a kind of rough division) conventionally.

3) take step 2) the trimap image that obtains is constraint, adopts image to take algorithm and extracts finger vena network clearly;

Image is taken off algorithm and is comprised two steps, estimates shade α algorithm and extracts foreground picture algorithm.

One, estimate shade α

The stingy drawing method of what the present invention taked is a kind of closed form extracts constraint image C (x, y), by constraint image C (x, y), solves following linear system and just can obtain shade α:

(L+λD s)α=λb s????(22)

Wherein, L scratches figure Laplace operator, and λ is a given larger numerical value, D sand b srepresent respectively a diagonal matrix and a vector that comprises constraint information.

Two, extract foreground picture

Carrying out above-mentioned estimation shade α step) afterwards, the extraction that we will complete finger vena network will be reconstructed F.Just can obtain a pair of suitable (F by minimizing following system *, B *):

&Psi; ( F * , B * ) = min &Sigma; i &Element; I ( ( &alpha; i F i + ( 1 - &alpha; i ) B i - I i ) 2 + | &alpha; i x | ( F i x 2 + B i y 2 ) + | &alpha; i y | ( F i y 2 + B i y 2 ) ) - - - ( 23 )

Wherein, be respectively F, B and α are for the derivative of x and y.So far, can extract foreground picture, obtain thus finger vena network clearly.

This method can accurately represent finger vena network, and can excavate thinner vein, and therefore, this method has great importance for the expression of finger vena network.

Claims (4)

1. the accurate extracting method of finger vena network, is characterized in that: the described accurate extracting method of finger vena network comprises the following step of carrying out in order:
1) original finger venous image is strengthened;
2) enhancing imagery exploitation automatic mark algorithm step 1) being obtained is processed, and obtains trimap image;
3) take step 2) the trimap image that obtains is constraint, adopts image to take algorithm and extracts finger vena network clearly.
2. the accurate extracting method of finger vena network according to claim 1, it is characterized in that: the method in described step 1), original finger venous image being strengthened is to utilize mist elimination Processing Algorithm to carry out scattered noise removal, obtain original mist elimination image R (x, y), strengthen the operator image E (x, y) that is enhanced through multiple dimensioned even Gabor filtering again.
3. the accurate extracting method of finger vena network according to claim 1, it is characterized in that: described step 2) in utilize the enhancing image that automatic mark algorithm obtains step 1) to process, the method that obtains trimap image is, the enhancing image E (x obtaining with step 1), y) be basis, utilize K-means to carry out clustering to strengthening image, successively generate the binary map of interim prospect and background by classification mark, recycle dynamic bianry image, generate the signature ζ of non-cross overlapping foreground pixel set through two-value operation operator fthe signature ζ of (x, y) and background pixel set b(x, y), afterwards with the signature ζ of foreground pixel set fthe signature ζ of (x, y) and background pixel set b(x, y) generates scribble district for constraining on original image, obtains retraining image C (x, y).
4. the accurate extracting method of finger vena network according to claim 1, it is characterized in that: in described step 3), utilize image to take algorithm to step 2) method extracted of the trimap image that obtains is first with constraint image C (x, y) estimate to obtain shade α, and then extract foreground picture as constraint by Optimized Iterative method take shade α, obtain thus finger vena network clearly.
CN201410102276.0A 2014-03-19 2014-03-19 Finger vein network accurate extracting method CN103903001A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201410102276.0A CN103903001A (en) 2014-03-19 2014-03-19 Finger vein network accurate extracting method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201410102276.0A CN103903001A (en) 2014-03-19 2014-03-19 Finger vein network accurate extracting method

Publications (1)

Publication Number Publication Date
CN103903001A true CN103903001A (en) 2014-07-02

Family

ID=50994313

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201410102276.0A CN103903001A (en) 2014-03-19 2014-03-19 Finger vein network accurate extracting method

Country Status (1)

Country Link
CN (1) CN103903001A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104933427A (en) * 2015-07-14 2015-09-23 中国民航大学 Embedded finger three-modal characteristic acquisition equipment

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090070860A1 (en) * 2007-09-06 2009-03-12 Hitachi, Ltd Authentication server, client terminal for authentication, biometrics authentication system, biometrics authentication method, and program for biometrics authentication
CN103336945A (en) * 2013-06-10 2013-10-02 黑龙江大学 Finger vein recognition method fusing local features and global features
CN103400134A (en) * 2013-06-28 2013-11-20 哈尔滨工程大学 Non-contact method for extracting region of interest of finger vein sample

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090070860A1 (en) * 2007-09-06 2009-03-12 Hitachi, Ltd Authentication server, client terminal for authentication, biometrics authentication system, biometrics authentication method, and program for biometrics authentication
CN103336945A (en) * 2013-06-10 2013-10-02 黑龙江大学 Finger vein recognition method fusing local features and global features
CN103400134A (en) * 2013-06-28 2013-11-20 哈尔滨工程大学 Non-contact method for extracting region of interest of finger vein sample

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
JINFENG YANG等: "finger-vein network enhancement and segmentation", 《PATTERN ANALYSIS AND APPLICATIONS》 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104933427A (en) * 2015-07-14 2015-09-23 中国民航大学 Embedded finger three-modal characteristic acquisition equipment

Similar Documents

Publication Publication Date Title
CN104392463B (en) Image salient region detection method based on joint sparse multi-scale fusion
Sadeghi et al. A novel method for detection of pigment network in dermoscopic images using graphs
Zhang et al. Learning a dilated residual network for SAR image despeckling
Miao et al. Road centerline extraction from high-resolution imagery based on shape features and multivariate adaptive regression splines
Du et al. Recognition of plant leaf image based on fractal dimension features
Lin et al. Line segment extraction for large scale unorganized point clouds
Bai et al. Splitting touching cells based on concave points and ellipse fitting
Liu et al. Iterative infrared ship target segmentation based on multiple features
CN102306290B (en) Face tracking recognition technique based on video
Fisher Cloud and cloud-shadow detection in SPOT5 HRG imagery with automated morphological feature extraction
CN100550978C (en) A kind of self-adapting method for filtering image that keeps the edge
CN105740758A (en) Internet video face recognition method based on deep learning
CN108573276B (en) Change detection method based on high-resolution remote sensing image
Celik A Bayesian approach to unsupervised multiscale change detection in synthetic aperture radar images
CN103136766B (en) A kind of object conspicuousness detection method based on color contrast and color distribution
Minhas et al. Shape from focus using fast discrete curvelet transform
CN103049763B (en) Context-constraint-based target identification method
CN106548463A (en) Based on dark and the sea fog image automatic defogging method and system of Retinex
CN105354865B (en) The automatic cloud detection method of optic of multispectral remote sensing satellite image and system
CN105760859A (en) Method and device for identifying reticulate pattern face image based on multi-task convolutional neural network
CN101650439B (en) Method for detecting change of remote sensing image based on difference edge and joint probability consistency
CN104268877B (en) A kind of infrared image sea horizon self-adapting detecting method
CN103942758A (en) Dark channel prior image dehazing method based on multiscale fusion
Zhang et al. Infrared small target detection via non-convex rank approximation minimization joint l2, 1 norm
CN104809734B (en) A method of the infrared image based on guiding filtering and visual image fusion

Legal Events

Date Code Title Description
PB01 Publication
C06 Publication
SE01 Entry into force of request for substantive examination
C10 Entry into substantive examination
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20140702