CN103559489A - Method for extracting features of palm in non-contact imaging mode - Google Patents

Abstract

The invention relates to a method for extracting the features of a palm in a non-contact imaging mode. The method includes the steps that firstly, an incircle of the palm at the roots of fingers is extracted, and a plurality of radial line segments drawn from the incenter are intersected at a circumference; meanwhile, a template is constructed to calculate gradient values of points in the incircle, the relative radius of a center of mass is calculated according to the gradient values of the points on the line segments, the relative radius is defined into a feature parameter, and feature vector space is constructed. The features constructed through the method have the advantages of being unchanged in translation, rotation and scaling and short in feature extraction time.

Description

Palm feature extracting method under a kind of noncontact imaging mode

Technical field:

The invention belongs to biometrics identification technology field, relate to palm feature extracting method under a kind of noncontact imaging mode.

Background technology:

Along with the development in epoch and social progress, gate control system more and more needs efficiently, ID authentication mechanism is determined the legitimacy that personnel pass in and out specific region reliably, and biometrics identification technology is a kind of technology of identifying identity according to the intrinsic physiological characteristic of human body or behavioural characteristic.Because biological characteristic has the feature of " everybody has, people is variant, remain unchanged for a long period of time ", and can not pass into silence or lose, be more and more applied in gate control system.Known according to existing pertinent literature, the biological characteristic that can be used at present gate control system comprises fingerprint, palmmprint, vein, people's face, iris, retina, people's ear etc.Hand-characteristic, because high and convenient while gathering image by consumers' acceptable degree, has obtained application widely in the middle of numerous biological characteristics.

Staff is at ordinary times in the state of partly clenching fist, and the feature in palm relatively points feature and the back of the hand feature is more difficult is stolen, and has higher security.Palm vein is the living body biological feature that is hidden in inside of human body, phase opponent's external texture feature, and vein pattern is more not easy to be stolen and copy.But because the physiological structure of palm self, the image medium sized vein information that some palm gathers under near infrared light almost estimate less than, the texture in palm is mainly comprised of palmmprint main line and some mastoid process lines.Therefore, use single collecting device herein, the palm to people under near infrared light is taken, and by the palm vein information that in the hand image obtaining, palm area comprises and palm palmmprint main line information definition, is palm feature, use it for identification, as shown in Figure 1.

The initial concept of palm vein identification appears at the nineties in 20th century, because of the popularization of Fujitsu's palm vein identification instrument, since 2006, has obtained broad research.At 2006 to 2010, hand collecting device all designed based on contact.Contact acquisition mode requires user staff and collecting device in gathering image process to come in contact, or holds some external units of collecting device or hand is placed on to fixedly fixedly fixing of putting position.Contact is captured in health and the problem of serving can be with in security aspect, simultaneously, the sensor surface of contact Acquisition Instrument is more easily contaminated, especially gate control system is often applied in the poor outdoor environment of sanitary condition, this can cause the rate of by mistake refusing of system to rise, and has also shortened the serviceable life of acquisition instrument simultaneously.

In 2010, start to occur the research for the palm characteristic recognition method of noncontact imaging mode.By cordless, realizing palm imaging can make hand identifier get the nod in the crowd who worries disease propagation, but the uncertain hand that causes of position exists translation, rotation and proportional zoom in image between hand and imaging device, it is the higher requirement of having extracted stable hand feature extraction.

Summary of the invention:

Goal of the invention:

The present invention relates to a kind of non-contact capture mode palm feature extracting method in image of setting about, its fundamental purpose is to improve the set about accuracy of identification of non-contact capture mode, hand translation in the hand image that solution causes because of the variation of position and angle between hand and imaging device, rotation and proportional zoom problem.

Technical scheme:

The present invention is achieved through the following technical solutions:

A palm feature extracting method under imaging mode, is characterized in that: the method step is as follows:

(1) selection of centre of the palm stable reference point: construct a centre of the palm incircle and select the stable reference point in the centre of the palm; Design incircle possesses following condition: tangent with palm both sides outline line respectively, and by the finger root point of middle finger, the third finger; To being positioned at outside forefinger the volar edge point on palm outline line on palm outline line and outside pinkie, scan respectively successively, determine position and the inscribe radius of a circle of incenter;

(2) extraction of palm invariant features: choose two kinds of palmar hand features of palm vein and palm palmmprint and carry out the realization of identification system; By incenter, obtain centre of the palm stable reference point, opponent's image positions, and eliminates the impact that gathers palm translation in image process; Point centered by the reference point of the centre of the palm, with angle θ ⁰for unit penetrates some rays, between centre of the palm reference point and incircle outline line, construct some radial line segments, be called Eigenvector; Finally, according to the gradient intensity value of each point on each Eigenvector, calculate the barycenter relative radius of line segment, structural attitude vector.

In above-mentioned steps (2), the concrete steps of structural attitude vector are as follows:

Step 1: the generation of gradient image

Select gradient intensity value as the basis of feature, before feature extraction, hand greyscale image transitions become to gradient image:

1) intersecting direction with palm vein or palm palmmprint main line, meet gray scale minimal value condition, therefore, by four direction (0 ⁰, 45 ⁰, 90 ⁰, 135 ⁰) judgement of gray scale minimal value, can find palm vein or palm palmmprint candidate pixel point; And do not meet the minimizing pixel of gray scale, be vein or palmmprint scarcely, therefore, by this pixel Grad zero setting;

2), in meeting the minimizing pixel of gray scale, within pixel is positioned at vein or palmmprint, at two Grad of its vertical direction, all will be larger; When pixel departs from vein or palmmprint, to there is small one and large one phenomenon in Grad, therefore, get smaller value in two Grad that meet gray scale minimal value direction as the Grad of this direction, get again the maximal value that meets in the minimizing all direction gradient values of gray scale as the Grad of this pixel, this Grad can reflect along vein or palmmprint vertical gradient variation tendency, be that maximum of gradients is corresponding to the center of vein or palmmprint, along with departing from central point, gradient reduces gradually;

Step 2: the extraction of Eigenvector center of mass point relative radius

From the zero degree direction of centre of the palm reference point, in the counterclockwise direction incircle is divided into H sector region, obtain thus H bar line segment K ₁, K ₂..., K _h; The distance of Eigenvector barycenter and centre of the palm reference point, barycenter radius, represents with r, the barycenter radius of j bar Eigenvector is:

$r_j=\frac{\underset{i=1}{\overset{M}{\mathrm{\Σ}}}i\×I_i}{\underset{i=1}{\overset{M}{\mathrm{\Σ}}}{I}_i}---\left(1\right)$

Wherein, Eigenvector is inscribed circle radius, and for convenience of calculating, by inscribed circle radius round numbers, the i in formula (1) is from 1 to M, and M is the incircle pixel number that radius comprises in the horizontal direction;

Except horizontal and vertical direction, unit point i on Eigenvector generally can not drop on pixel, its Grad adopts interpolation method to calculate, and adopts bilinear interpolation method, utilizes the Grad of 4 nearest neighbor pixels of unit point to carry out the Grad of unit of account point;

In order to solve proportional zoom problem, the relative radius δ to center of mass point and incenter _jcalculate, computing formula is as follows:

${\mathrm{\δ}}_j=\frac{r_j}{R}---\left(2\right)$

Utilize the relative radius of H bar Eigenvector center of mass point and incenter to construct the invariant features under translation, rotation and proportional zoom conversion, proper vector is:

F＝{δ ₁,δ ₂,…,δ _H}；

Step 3: the coupling of proper vector

Through above-mentioned steps, in palm, construct the line segment that H bar can reflect palm textural characteristics, establish the proper vector extracted from register hand image as:

F _a＝{δ _1a,δ _2a,…,δ _Ha}，

The proper vector of extracting from hand image to be identified is:

F _b＝{δ _1b,δ _2b,…,δ _Hb}，

Judge whether registration hand image mates with hand image to be identified, first judges that the two has how many coupling line segments pair, and in mating, the proper vector between calculated characteristics line segment is apart from d at line segment _jas the similarity criteria of corresponding line segment barycenter relative radius in 2 width images, the proper vector of registration hand image and hand image middle conductor j to be identified is as follows apart from computing formula:

d _j＝|δ _ja-δ _jb| (3)

Work as d _jbe less than setting threshold T ₁time, think that two corresponding line segments are for coupling line segment, if coupling line segment is to there being g, matching rate is:

$G=\frac{g}{H}\×100\%---\left(4\right)$

When matching rate G is greater than setting threshold T ₂time, think that two width hand images derive from same person.

Advantage and effect:

The feature that the present invention constructs has translation, rotation and proportional zoom unchangeability, and meanwhile, the feature extracting method travelling speed providing is very fast, is more suitable for the noncontact hand recognition device higher to requirement of real-time.

Accompanying drawing explanation:

Fig. 1 is the palm picture collecting under cordless;

Fig. 2 is that palm incircle detects schematic diagram;

Fig. 3 is palm feature extracting method process flow diagram under noncontact imaging mode;

Fig. 4 is gradient value schematic diagram;

Fig. 5 is gradient calculation template, and Fig. 5 (a) is gradient calculation template schematic diagram, Fig. 5 (b) for l be 2 o'clock subtemplate center position schematic diagram, Fig. 5 (c) for l be 3 o'clock subtemplate center position schematic diagram;

Fig. 6 is the palm invariant features structural map based on vena metacarpea and palmmprint specified point position, Fig. 6 (a) is structural attitude line chart in incircle, Fig. 6 (b) is Eigenvector schematic diagram, Fig. 6 (c) is pixel gray-scale value schematic diagram on Eigenvector, and Fig. 6 (d) is Eigenvector center of mass point schematic diagram.

Embodiment:

Below in conjunction with accompanying drawing and concrete embodiment, the present invention is described further:

The present invention is palm feature extracting method under a kind of noncontact imaging mode, it is characterized in that: while utilizing noncontact imaging mode to gather hand image, palmistry is uncertain to the direction of imaging device, position and shooting distance, this causes for same person, and in each hand image gathering, size, direction and the position of hand may be not identical.The feature that the direct extraction of this method of the present invention and translation, rotation and proportional zoom are irrelevant, has avoided in process, may changing the some properties of original image and having increased working time being normalized in image.The method step is as follows:

(1) selection of centre of the palm stable reference point: for the irrelevant feature of relative space position structure and palm rotation, translation and proportional zoom of utilizing each point and centre of the palm stable reference point on palm vein and palmmprint main line, select the stable reference point in the centre of the palm by constructing a centre of the palm incircle; In order to guarantee that incircle has uniqueness, design incircle possesses following condition: tangent with palm both sides outline line respectively, and by the finger root point of middle finger, the third finger; From the geometric properties of justifying, on round edge circle, the perpendicular bisector of any two points must be through the center of circle, and the tangent line of circle must be perpendicular to the radius through this point of contact, according to these two character, to being positioned at outside forefinger the volar edge point on palm outline line on palm outline line and outside pinkie, scan respectively successively, determine position and the inscribe radius of a circle of incenter; Step is as follows:

1) utilize the hand shape contour tracing method based on direction gradient extreme value to extract palm outline line.

2) extraction middle finger, the third finger refer to root point P ₁concrete methods of realizing is as follows: for left hand imaging, its thumb in image upward, the third finger is positioned under middle finger, from middle fingertip, along step 1) outline line that extracts finds first curvature to change point greatly, this point is middle finger and nameless finger tip intersection C, in corresponding gray level image, from C point, use gray scale minimal value tracking to find middle finger and nameless webs line again; The arbitrary pixel being set on webs line is P ₁₁, find its two some P in vertical webs direction ₁₂and P ₁₃, these 2 and some P ₁₁distance be that 3 pixels are wide; Difference calculation level P ₁₁with a P ₁₂, some P ₁₃grad, using the mean value AVG of two Grad as evaluation point P ₁₁graded amount in vertical webs direction; The C point of take is starting point, and each pixel being positioned on webs line is calculated to its graded amount AVG successively, and when AVG occurs obviously to reduce, tracking stops, and the more front finger root point P1 that is middle finger, the third finger, is marked at this point in handwheel profile bianry image.

3) refer to that root incircle extracts, palm incircle automatic testing method specific implementation step is as follows: a) determine the scope at incircle and palm outline line point of contact, the marginal point within the scope of this is as sweep object.First, by the finger root point P of middle finger, the third finger ₁be straight line L ₃, this straight line is perpendicular to the webs fitting a straight line between middle finger and the third finger.Straight line L ₃intersect at palm outline line, taking off first point that side starts is P ₂, second point is P ₃.Secondly, from a P ₃beginning is made series of parallel line to palm wrist direction, and parallel lines are parallel to line L ₃, every parallel lines intersect at outline line, take off second point that side starts, and form some series, and when the ordinate of point occurs to jump, getting this point is P ₄.Finally, construct two some set A ₁and A ₂.First set A ₁by from a P ₃to a P ₄between edge line on point form; Second some set A ₂by wrist marginal point Q to some a P ₂between edge line on point form.Point set A ₁and A ₂the scope that is incircle and palm outline line point of contact, the marginal point within the scope of this, as sweep object, performs step respectively b) and step c).B) extract incenter candidate point region.First, choose set A ₁or set A ₂in any point, be made as a P ₅.By the finger root point P of middle finger, the third finger ₁with marginal point P ₅do line, establishing line segment is L ₂, to line segment L ₂be perpendicular bisector L ' ₂; Secondly, along palm outline line at a P ₅2 points are respectively got in left and right, by these points, are fitting a straight line L ₁, cross P ₅point is straight line L ₁vertical line L ' ₁.Straight line L ' ₁with straight line L ' ₂intersect at an O ₁.Pair set A ₁and A ₂interior each point is made respectively aforesaid operations, extracts its joining O ₁, obtain respectively a set A ' ₁and A ' ₂.Set of computations A ' ₁in point and A ' ₂the Euclidean distance of mid point, the point that distance is less than to Δ A retains, form set A ' (the present invention gets Δ A=6), in pair set A ', horizontal stroke, the ordinate of each point compare, and obtain minX, minY, maxX, maxY, forms the candidate point that in ，Gai rectangular area, a rectangular area, each pixel is incenter.C) determine incenter and inscribed circle radius.First calculate in candidate point region, the center of circle outside each center of circle candidate point and forefinger the distance of palm outline line outside palm outline line and pinkie, computing method are as follows: choose the arbitrary center of circle candidate point in candidate point region, the center of circle, be made as O ', calculate respectively center of circle candidate point O ' and set A ₁, A ₂the Euclidean distance of middle each point, gets minimum value, is made as d ₁and d ₂, apart from d ₁and d ₂be pixel O ' to the distance of both sides palm outline line.Then calculated candidate point O ' refers to root point P with middle finger, the third finger ₁distance be d ₃.Finally calculate apart from d ₁, d ₂and d ₃standard deviation, evaluate the degree that three distances differ.Each center of circle candidate point is done respectively to aforesaid operations, select minimumly in the standard deviation series obtaining, corresponding candidate point is incenter O (x ₀, y ₀), this point and middle finger, nameless and palm interfaces point P ₁distance be inscribed circle radius R.D) with an O (x ₀, y ₀) be the center of circle, R is radius, draws palm incircle, rejects circle external information, is palm vein feature extraction region.

(2) extraction of palm invariant features: in the hand image gathering under near infrared light, palm vein presents netted, present native texture state together with palm palmmprint main line, therefore a width hand image can be regarded a width texture image as, and palm vein is the important component part of palm textural characteristics.But because the physiological structure of palm self, the image medium sized vein information that some palm gathers under near infrared light almost estimate less than, the texture in palm is mainly comprised of palmmprint main line and some mastoid process lines.Palmmprint main line and mastoid process line not only have very strong directivity, and thickness is deep mixed, and therefore, the present invention chooses palm vein and two kinds of palmar hand features of palm palmmprint are carried out the realization of identification system; By incenter, obtain centre of the palm stable reference point, opponent's image positions, and eliminates the impact that gathers palm translation in image process; Point centered by the reference point of the centre of the palm, with angle θ ⁰for unit penetrates some rays, between centre of the palm reference point and incircle outline line, construct some radial line segments (line segment length equals inscribed circle radius length), be called Eigenvector; Finally, according to the gradient intensity value of each point on each Eigenvector, calculate the barycenter relative radius (being the absolute distance of barycenter and incenter and the ratio of inscribed circle radius) of line segment, structural attitude vector.

In above-mentioned steps (2), the concrete steps of structural attitude vector are as follows:

Step 1: the generation of gradient image

The extraction not a duck soup of palm vein and palmmprint locus, even if less mistake also can be brought larger error to identification.For this reason, the present invention selects gradient intensity value as the basis of feature, before feature extraction, hand greyscale image transitions is become to gradient image, and transfer principle is as follows:

1) intersecting direction with palm vein or palm palmmprint main line, meet gray scale minimal value condition, therefore, by four direction (0 ⁰, 45 ⁰, 90 ⁰, 135 ⁰) judgement of gray scale minimal value, can find palm vein or palm palmmprint candidate pixel point; And do not meet the minimizing pixel of gray scale, be vein or palmmprint scarcely, therefore, by this pixel Grad zero setting;

2), in meeting the minimizing pixel of gray scale, within pixel is positioned at vein or palmmprint, at two Grad of its vertical direction, all will be larger; When pixel departs from vein or palmmprint, to there is small one and large one phenomenon in Grad, therefore, get smaller value in two Grad that meet gray scale minimal value direction as the Grad of this direction, get again the maximal value that meets in the minimizing all direction gradient values of gray scale as the Grad of this pixel, this Grad can reflect along vein or palmmprint vertical gradient variation tendency, be that maximum of gradients is corresponding to the center of vein or palmmprint, along with departing from central point, gradient reduces gradually;

Step 2: the extraction of Eigenvector center of mass point relative radius

From the zero degree direction of centre of the palm reference point, in the counterclockwise direction incircle is divided into H sector region, obtain thus H bar line segment K ₁, K ₂..., K _h; The distance of Eigenvector barycenter and centre of the palm reference point, barycenter radius, represents with r, the barycenter radius of j bar Eigenvector is:

$r_j=\frac{\underset{i=1}{\overset{M}{\mathrm{\Σ}}}i\×I_i}{\underset{i=1}{\overset{M}{\mathrm{\Σ}}}{I}_i}---\left(1\right)$

Wherein, Eigenvector is inscribed circle radius, and for convenience of calculating, by inscribed circle radius round numbers, the i in formula (1) is from 1 to M, and M is the incircle pixel number that radius comprises in the horizontal direction;

Except horizontal and vertical direction, unit point i on Eigenvector generally can not drop on pixel, its Grad adopts interpolation method to calculate, and the present invention adopts bilinear interpolation method, utilizes the Grad of 4 nearest neighbor pixels of unit point to carry out the Grad of unit of account point;

Because hand imaging is not of uniform size, the inscribed circle radius calculating is not unique, and the distance in the relative center of circle of center of mass point is whipped imaging size and changed, in order to solve proportional zoom problem, and the relative radius δ to center of mass point and incenter _j(be the barycenter radius r of center of mass point and incenter _jratio with inscribed circle radius R) calculate, computing formula is as follows:

${\mathrm{\δ}}_j=\frac{r_j}{R}---\left(2\right)$

Utilize the relative radius of H bar Eigenvector center of mass point and incenter to construct the invariant features under translation, rotation and proportional zoom conversion, proper vector is:

F＝{δ ₁,δ ₂,…,δ _H}；

Step 3: the coupling of proper vector

Through above-mentioned steps, can in palm, construct the line segment that H bar can reflect palm textural characteristics, establish the proper vector extracted from register hand image as:

F _a＝{δ _1a,δ _2a,…,δ _Ha}，

The proper vector of extracting from hand image to be identified is:

F _b＝{δ _1b,δ _2b,…,δ _Hb}，

Judge when whether registration hand image mates with hand image to be identified, need to first judge that the two has how many coupling line segments pair, at line segment, in mating, the proper vector between calculated characteristics line segment is apart from d _jas the similarity criteria of corresponding line segment barycenter relative radius in 2 width images, the proper vector of registration hand image and hand image middle conductor j to be identified is as follows apart from computing formula:

d _j＝|δ _ja-δ _jb| (3)

Work as d _jbe less than setting threshold T ₁time, think that two corresponding line segments are for coupling line segment, if coupling line segment is to there being g, matching rate is:

$G=\frac{g}{H}\×100\%---\left(4\right)$

When matching rate G is greater than setting threshold T ₂time, think that two width hand images derive from same person.

The present invention relates to palm feature extracting method under a kind of noncontact imaging mode, in opponent's image, palm localization method and palm feature extracting method are studied, its fundamental purpose is to improve the set about accuracy of identification of non-contact capture mode, hand translation in the hand image that solution causes because of the variation of position and angle between hand and imaging device, rotation and proportional zoom problem.

When the present invention is mated with the hand image of cordless collection, hand need to be normalized, thereby translation, rotation and proportional zoom that palm feature and hand imaging are caused are irrelevant.Because normalization process may change the some properties of original image, and the extraction that this part character changing may effect characteristics, meanwhile, normalization process also can increase the time of operation.Therefore, the present invention selects directly to have constructed palm feature a kind of and that translation, rotation and proportional zoom are irrelevant.

By reference to the accompanying drawings concrete steps are described as follows:

(1) Fig. 1 is the palm image gathering by cordless under near infrared light.By the palm image gathering under the near infrared light to a large amount of, observe, in palm image, can clearly see palmmprint main line information, can utilize same collecting device to extract palm vein and palmmprint main line completely.From same width hand image, extract vena metacarpea feature and the palmmprint main line feature of staff, can avoid the problem that some multi-modal biological characteristics merge, as system complexity, cost, user's acceptance, data base administration etc.

(2) Fig. 2 is that palm incircle detects schematic diagram.For contactless hand recognition system, the staff image collecting is often subject to the impact of the factors such as distance of camera lens, hands movement, misoperation, causes the very large difference of existence between image.For the impact of removal of images difference on recognition effect, from palm, extract suitable reference point, set up new reference frame, opponent's image positions, to reduce the impact of the factors such as the rotation introduced in image acquisition process, translation, proportional zoom, improve the robustness of coupling recognizer.

(3) Fig. 3 is palm feature extracting method process flow diagram under noncontact imaging mode, and this method proposes a kind of palm feature that can reflect palm arteries and veins and palm print characteristics and not be subject to rotate and proportional zoom affects.First extract the palm incircle that refers to root, from incenter, do some radial line segments and intersect at circumference, simultaneously, the Grad of each point in structure formwork calculation incircle, according to the Grad of each point on each line segment, calculate barycenter relative radius, be defined as characteristic parameter, with this constitutive characteristic vector space.

(4) Fig. 4 is gradient value schematic diagram.In palm image, intersecting direction with palm vein or palm palmmprint main line, meet gray scale minimal value condition, therefore, by four direction (0 ⁰, 45 ⁰, 90 ⁰, 135 ⁰) judgement of gray scale minimal value, can find palm vein or palm palmmprint candidate pixel point.And do not meet the minimizing pixel of gray scale, be vein or palmmprint scarcely, therefore, by this pixel Grad zero setting.In meeting the minimizing pixel of gray scale, within pixel is positioned at vein or palmmprint, at two Grad of its vertical direction, all will be larger; When pixel departs from vein or palmmprint, to there is small one and large one phenomenon in Grad, therefore, get smaller value in two Grad that meet gray scale minimal value direction as the Grad of this direction, get again the maximal value that meets in the minimizing all direction gradient values of gray scale as the Grad of this pixel, this Grad can reflect along vein or palmmprint vertical gradient variation tendency, be that maximum of gradients is corresponding to the center of vein or palmmprint, along with departing from central point, gradient reduces gradually.

(5) Fig. 5 is gradient calculation template.Utilize the gradient template of structure hand greyscale image transitions can be become to gradient image.

(6) Fig. 6 is the schematic diagram of the palm invariant features structure based on palmmprint and vena metacarpea specified point position.The basic thought of palm latent structure is as follows: between centre of the palm reference point and incircle outline line, construct some radial line segments, every line segment may be through some palm veins or palmmprint streakline.When the line segment with abundant removes to cover the hand image in incircle, and just the distributed architecture of palmmprint and vena metacarpea in the hand image in incircle can be described out with these line segments in conjunction with corresponding grey scale change information.

Known according to the constant principle of moment of single order, Eigenvector each point Grad and this sum of products of putting relative centre of the palm reference point distance equal the product of this Eigenvector each point Grad sum centre of the palm relative to center of mass point reference point distance, thus, can utilize the distance of the Grad acquisition relative centre of the palm of the Eigenvector center of mass point reference point of each point on Eigenvector, be defined as the relative radius of center of mass point.

In hand image, palm vein has larger gray scale sudden change with relative non-streakline place, palmmprint main line place, has higher Grad, and the impact of line segment centroid position is played a decisive role.Therefore, can utilize the barycenter relative radius of Eigenvector to carry out the grey scale change of each point on Expressive Features line segment, and then embody the distributed architecture of vena metacarpea and palmmprint in the hand image in incircle.

Select the relative radius of line segment barycenter and incenter as palm feature, this feature is guaranteed hand image scaled convergent-divergent unchangeability.Select centre of the palm reference point to middle finger, the third finger, to refer to that the line direction of root point is zero degree direction, the direction using it as article one Eigenvector.Position, direction and the size of due to the position of this point, not whipping in image change, and therefore, the direction of article one Eigenvector has uniqueness.According to above analysis, every Eigenvector barycenter relative radius can reflect that palm vein and palmmprint are in the position distribution in hand space, usings that this has translation as palm feature, rotation and proportional zoom unchangeability.

Claims (2)

1. a palm feature extracting method under noncontact imaging mode, is characterized in that: the method step is as follows:

(1) selection of centre of the palm stable reference point: construct a centre of the palm incircle and select the stable reference point in the centre of the palm; Design incircle possesses following condition: tangent with palm both sides outline line respectively, and by the finger root point of middle finger, the third finger; To being positioned at outside forefinger the volar edge point on palm outline line on palm outline line and outside pinkie, scan respectively successively, determine position and the inscribe radius of a circle of incenter;

(2) extraction of palm invariant features: choose two kinds of palmar hand features of palm vein and palm palmmprint and carry out the realization of identification system; By incenter, obtain centre of the palm stable reference point, opponent's image positions, and eliminates the impact that gathers palm translation in image process; Point centered by the reference point of the centre of the palm, with angle θ ⁰for unit penetrates some rays, between centre of the palm reference point and incircle outline line, construct some radial line segments, be called Eigenvector; Finally, according to the gradient intensity value of each point on each Eigenvector, calculate the barycenter relative radius of line segment, structural attitude vector.

2. palm feature extracting method under noncontact imaging mode according to claim 1, is characterized in that: in step (2), the concrete steps of structural attitude vector are as follows:

Step 1: the generation of gradient image

Select gradient intensity value as the basis of feature, before feature extraction, hand greyscale image transitions become to gradient image:

1) intersecting direction with palm vein or palm palmmprint main line, meet gray scale minimal value condition, therefore, by four direction (0 ⁰, 45 ⁰, 90 ⁰, 135 ⁰) judgement of gray scale minimal value, can find palm vein or palm palmmprint candidate pixel point; And do not meet the minimizing pixel of gray scale, be vein or palmmprint scarcely, therefore, by this pixel Grad zero setting;

2), in meeting the minimizing pixel of gray scale, within pixel is positioned at vein or palmmprint, at two Grad of its vertical direction, all will be larger; When pixel departs from vein or palmmprint, to there is small one and large one phenomenon in Grad, therefore, get smaller value in two Grad that meet gray scale minimal value direction as the Grad of this direction, get again the maximal value that meets in the minimizing all direction gradient values of gray scale as the Grad of this pixel, this Grad can reflect along vein or palmmprint vertical gradient variation tendency, be that maximum of gradients is corresponding to the center of vein or palmmprint, along with departing from central point, gradient reduces gradually;

Step 2: the extraction of Eigenvector center of mass point relative radius

From the zero degree direction of centre of the palm reference point, in the counterclockwise direction incircle is divided into H sector region, obtain thus H bar line segment K ₁, K ₂..., K _h; The distance of Eigenvector barycenter and centre of the palm reference point, barycenter radius, represents with r, the barycenter radius of j bar Eigenvector is:

$r_j=\frac{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}i\×I_i}{\underset{i=1}{\overset{M}{\mathrm{\Σ}}}{I}_i}---\left(1\right)$

Wherein, Eigenvector is inscribed circle radius, and for convenience of calculating, by inscribed circle radius round numbers, the i in formula (1) is from 1 to M, and M is the incircle pixel number that radius comprises in the horizontal direction;

Except horizontal and vertical direction, unit point i on Eigenvector generally can not drop on pixel, its Grad adopts interpolation method to calculate, and adopts bilinear interpolation method, utilizes the Grad of 4 nearest neighbor pixels of unit point to carry out the Grad of unit of account point;

In order to solve proportional zoom problem, the relative radius δ to center of mass point and incenter _jcalculate, computing formula is as follows:

${\mathrm{\δ}}_j=\frac{r_j}{R}---\left(2\right)$

Utilize the relative radius of H bar Eigenvector center of mass point and incenter to construct the invariant features under translation, rotation and proportional zoom conversion, proper vector is:

F＝{δ ₁,δ ₂,…,δ _H}；

Step 3: the coupling of proper vector

Through above-mentioned steps, in palm, construct the line segment that H bar can reflect palm textural characteristics, establish the proper vector extracted from register hand image as:

F _a＝{δ _1a,δ _2a,…,δ _Ha}，

The proper vector of extracting from hand image to be identified is:

F _b＝{δ _1b,δ _2b,…,δ _Hb}，

Judge whether registration hand image mates with hand image to be identified, first judges that the two has how many coupling line segments pair, and in mating, the proper vector between calculated characteristics line segment is apart from d at line segment _jas the similarity criteria of corresponding line segment barycenter relative radius in 2 width images, the proper vector of registration hand image and hand image middle conductor j to be identified is as follows apart from computing formula:

d _j＝|δ _ja-δ _jb| (3)

Work as d _jbe less than setting threshold T ₁time, think that two corresponding line segments are for coupling line segment, if coupling line segment is to there being g, matching rate is:

$G=\frac{g}{H}\×100\%---\left(4\right)$

When matching rate G is greater than setting threshold T ₂time, think that two width hand images derive from same person.

Images