CN101055618A - Palm grain identification method based on direction character - Google Patents

Abstract

The invention discloses a method for identifying the palmprint based on directional characteristic, including the steps of: collecting the palmmprint image, preprocessing palmmprint image, constructing the palmmprint trainning ROI image collection, extracting-establishing the palmmprint direction feature coding RPOC by the palmmprint feature, matching the palmmprint in the dot pairs area. First new palmmprint trainning ROI image is established by rotating the palmmprint trainning ROI image and is added into the palmmprint trainning ROI image collection for compensating the rotation error; secondly, a modified limited Radon conversion MFRAT is designed, and the palmmprint direction characteristic mode is obtained by comparing the energy sizes of linear areas in six direction at the local area of palmmprint image; Finally, the palmmprint direction characteristic mode is matched through the match of dot pair area. The palmmprint direction feature coding RPOC in the method, not only can reflect the structural feature of palmprint, also can carry major distinguishing information, at the same time, having good fault-tolerant ability for light change, displacement revolving between palmmprint images or the like.

Description

Based on directional characteristic palm grain identification method

Affiliated field the present invention relates to a kind of method of utilizing human body biological characteristics to carry out authentication, and is particularly a kind of based on directional characteristic palm grain identification method.

In the background technology network information society, an urgent demand can effectively be differentiated people's true identity.Traditional authentication identifying method mainly contains two kinds: one is based on the security mechanism of password; Two are based on the security mechanism of certificate.But these traditional affirmation mechanism have its intrinsic drawback.For example, a plurality of passwords are difficult to memory and forget easily; Certificate is forged easily, is stolen and lose etc.So people wait in expectation and use safer, more convenient identification authentication mode.

In recent ten years, the identity identifying technology (Biometrics) based on human body biological characteristics more and more is subject to people's attention.So-called biometrics identification technology be meant utilize human body itself intrinsic physical features or behavioural characteristic, differentiate the technology of personal identification by methods such as Flame Image Process, pattern-recognitions.Compare with traditional identification authentication mode based on password or ID card, it can carry, be difficult to forge and need not remember, and therefore has better security, reliability and validity.At present, the physical features of studying in the biometrics identification technology mainly contains fingerprint, people's face, iris, palmmprint, hand shape, ear line and vein etc.; The behavioural characteristic of research mainly contains person's handwriting, sound, gait and keystroke etc.In above-mentioned research, differentiate that based on the identity of fingerprint being the earliest, also is the most ripe a kind of method, but the further popularization that the property easy to wear of fingerprint and destructible have limited this method to a certain extent.Though the authentication identifying method based on iris and cornea has the discrimination advantages of higher, also exist defectives such as apparatus expensive, user's acceptance difference.People's face and sound also are the research focuses of biometrics identification technology, but because the influence of factors such as the attitude of illumination condition, people's face, noise, its accuracy rate also is difficult to satisfactory

People's palmmprint has uniqueness and lifelong constant substantially characteristic, compares with fingerprint, and palmmprint regional big have abundanter texture information more, and obtaining also of palmprint image is more prone to.Recent years, the biometrics identification technology research of discerning based on palmmprint also has been subjected to extensive concern.

Early stage palmmprint identification is off-line processing, gathers palmmprint and is to use ink to restrain palm on blank sheet of paper, uses digital camera, scanner etc. to obtain digital picture then.Yet the identification of off line palmmprint is applied to have a lot of shortcomings in the authentication work.At first, it is not a real-time, and most practical application be requirement can online in real time identification; Secondly, the printing ink palmprint image is of low quality, and feature can't be by high-quality performance; Once more, in the off line image, the ROI of palmmprint (Regionof Interest) zone is difficult to effective location.After 2002, the palmmprint Study of recognition is transferred to online palmprint gradually and is discerned up in the world.Online palmprint identification just is to use digital equipment directly to obtain the better quality palmprint image with the difference of off line palmmprint identification maximum, can handle in real time.Wherein, be in the relevant paper of purpose with research, the common equipment that obtains palmprint image is CCD camera, digital scanner etc.Here it is to be noted scheming in The Hong Kong Polytechnic University's living things feature recognition especially the palmmprint of a special use obtain equipment, obtained the 14th show of inventions gold medal of China.

In international and domestic paper data Kuku, can retrieve the paper of more than 100 piece of palmmprint identification aspect.Generally speaking, the palm grain identification method that is proposed can be divided into following a few class:

(1) based on the method for textural characteristics.These class methods are regarded palmprint image as a kind of texture structure, and the textural characteristics that uses correlation technique to extract palmmprint is discerned.The PalmCode of proposition such as D.Zhang and A.Kong is a kind of palm grain identification method based on texture of classics, it uses the Gabor wave filter that palmprint image is carried out filtering, use the zero crossing criterion then to the palmprint image [list of references: D.Zhang of encoding, W.K.Kong, J.You, and M.Wong, " Onlinepalmprint identification; " IEEE Transactions on Pattern Analysis and MachineIntelligence, 25 (9) (2003), pp.1041-1050.].Use information fusion technology such as A.Kong has carried out improving to PalmCode and has proposed FusionCode subsequently, discrimination [the list of references: A.Kong that is further improved, D.Zhang, and M, Kamel, " Palmprint identificationusing feature-level fusion, " Pattern Recognition.39 (2006) 478-487].

(2) based on the method for line feature.L.Zhang etc. at first carry out wavelet decomposition to palmprint image, the service orientation modeling method is extracted the significant coefficient of wavelet sub-band as main line and important THE FOLD FEATURES [list of references: L.Zhang then, and D.Zhang, " Characterization of palmprints bywavelet signatures via directional context modeling; " IEEE Transaction onSystems, Man and Cybernetics, Part is (3) (2004) B.34, pp.1335-1347.].Wu Xiangqian etc. then regard palm line as a kind of ridge line, first order derivative zero crossing according to image, the amplitude of second derivative is determined palm line, on the basis of this algorithm, Wu Xiangqian etc. have inquired into based on the classification of the palmmprint of line feature and recognizer [list of references: X.Q.Wu, D.Zhang, and K.Q.Wang, " Palm line extraction and matching for personal authentication; " IEEETransaction on Systems, Man and Cybernetics, Part A, 36 (5) (2006), pp.978-987.].

(3) based on the method for characteristic feature or be called subspace method.Technology such as main use characteristic value decomposition and Gram-Schmidt orthogonalization are carried out dimensionality reduction to palmprint image, and obtain the individual features vector.G.M.Lu and X.Q.Wu propose the palm grain identification method [list of references: G.M.Lu with LDA (linear discriminant analysis) based on PCA (principal component analysis (PCA)) respectively, D.Zhang, and K.Q.Wang, " Palmprint recognition using eigenpalms features; " PatternRecognition Letter, 24 (9-10) (2003), pp.1463-1467.], [list of references: X.Q.Wu, D.Zhang, and K.Q Wang, " Fisherpalms based palmprint recognition; " PatternRecognitio Letter, 24 (1 5) (2003), pp.2829-2838.].

(4) based on the method for directional information.The method is similar to the field of direction of fingerprint image and calculates and estimation.For palmprint image, try to achieve the direction of each pixel of palmprint image exactly, thereby palmprint image is mapped to the direction character space from gray space, mate then.In palmmprint identification field, obtained higher discrimination at present based on directional characteristic algorithm, because the directional information of palmmprint line can be carried more identifying information, and also insensitive to variation situations such as illumination variation.A.Kong and David.Zhang (magnifying roc) propose CompetitiveCode on the FusionCode basis, in the method, use the Gabor wave filter of 6 directions that image is carried out filtering, use the strongest response direction of Winner-take-all Rule Extraction as recognition feature [list of references: A.Kong then, and D.Zhang, " Competitive coding scheme for palmprint verification, " Proc.Of the 17 ^ThICPR, vol (1) (2004), pp.520-523.].

In above-mentioned several recognition methods, the first kind is the comparison morning of research based on the method for texture, but this method is subjected to the influence of factors such as illumination variation easily, and discrimination is difficult to improve.Second class then is subject to many limitations based on the method for line, and for example, many palm lines are relatively fuzzyyer to be difficult to extract.The 3rd class is the research focus based on the recognition methods of sub-space technique at present, but this method is only considered the correlativity between image pixel, does not utilize the structural information of palmprint image, and the recognition result on fairly large database awaits further to verify.

Comparatively speaking, the 4th class then can obtain comparatively ideal recognition result based on directional characteristic palmprint image recognition methods.At present, in the comparison of the whole bag of tricks recognition result, best based on directional characteristic method discrimination.The classical CompetitiveCode[list of references that is exactly above us to be mentioned based on directional characteristic palm grain identification method: A.Kong, and D.Zhang, " Competitivecoding scheme for palmprint verification, " Proc.Of the 17 ^ThICPR, vol (1) (2004), pp.520-523.].

By the international monopoly database retrieval, The Hong Kong Polytechnic University magnifies at the living things feature recognition center roc people such as (David.Zhang) and has applied for U.S.'s patent of invention in June, 2004, its patent announcement number is WO/2005/124662, and name is called " based on the directional characteristic palmmprint discrimination method of palm line ".The core content of the palmmprint discrimination method of this disclosure of the Invention is exactly CompetitiveCode.

Yet, the some shortcomings part is still arranged based on the palm grain identification method of CompetitiveCode.Be in particular in: the Gabor wave filter that uses in the palm grain identification method of (1) CompetitiveCode is not to extract the directional characteristic best instrument of palmprint image; (2) feature extraction speed is slow, because use more consuming time to palmprint image filtering than the Gabor wave filter of large form; (3) do not have the solution of proposition at the rotation problem, discrimination is difficult to further raising; (4) palm grain identification method of CompetitiveCode use Hamming distance is mated and is lacked fault-tolerant ability preferably.

Summary of the invention the objective of the invention is to overcome the deficiency in the palm grain identification method of prior art CompetitiveCode, proposes a kind of new for directional characteristic palm grain identification method.Not only have very fast feature extraction speed based on directional characteristic palm grain identification method, and discrimination is greatly improved also.This method has stronger robustness and better practicality than the palm grain identification method of CompetitiveCode.

The object of the present invention is achieved like this: based on directional characteristic palm grain identification method, comprise

Step (1) palm-print image capture

Harvester by palmprint image is gathered palmprint image, obtains can be used for the further palmprint image gray matrix of processing.Palmprint image in the registration phase collection is called the palmmprint training image, is called the palmmprint test pattern at the palmprint image of cognitive phase collection.

Step (2) palmprint image pre-service

Before extracting the palmprint image feature, need carry out pre-service to palmprint image.When gathering palmprint image, generally be the palmprint image of gathering whole palm, be inappropriate but be to use this palmprint image to mate, because the palmprint image of whole palm is bigger, processing speed is slow, is not suitable for real-time application on the one hand; Because the palmprint image of whole palm does not pass through localization process, there be very big rotation, displacement error on the other hand, make the matching result instability.Therefore in the palmprint image recognition methods, at first by location palm, finger position, palmprint image is rotated correction, train ROI (Region of Interest) image in the square region of the centre of palmprint image cutting 128 * 128 pixels as palmmprint then, the palmmprint training ROI image in last square shaped zone carries out feature extraction and coupling.

Particularly also comprise: the structure of step (3) palmmprint training plan image set

In Palm Print Recognition System, at first can gather a width of cloth or a few width of cloth palmprint image leaves in the system as the palmmprint training image.Yet because incomplete pretreatment operation, often there is certain rotation error in palmprint image to be identified with the palmmprint training image, and identification easily makes the mistake.At present, palmmprint identification field does not also propose the effective scheme of head it off.The present invention uses the new palmmprint training image of structure to solve this problem.By observing, the maximum rotation error between palmprint image to be identified and palmmprint training image is about 10 °.If palmmprint to the someone, there is width of cloth palmmprint training ROI image A in the system, to palmmprint training ROI image A respectively anglec of rotation α be 3 °, 6 °, 9 ° ,-3 ° ,-6 ° ,-9 °, obtain postrotational palmmprint training ROI image, promptly form new palmmprint training ROI image set A1, A2, A3, A4, A5, A6, set up departments and also include several palmmprints of B, C training ROI image in the system, then last palmmprint training ROI image set is A, A1, A2, A3, A4, A5, A6, B, C, structure by palmmprint training ROI image set can effectively compensate rotation error;

Anglec of rotation α wherein, the quantity of the new palmmprint training image of generation can be according to the actual conditions adjustment.

Step (4) palm print characteristics extracts-sets up palmmprint direction character coding RPOC

The principal character of palmmprint is the line feature, and simultaneously these lines have directivity, so the essential structure that direction character also can the effective expression palmmprint.At the regional area of palmprint image, therefore the palmmprint line can be used limited Radon conversion (the finiteRadon transform can abbreviate FRAT as) to calculate the direction of palm line by the approximate short lines of regarding as.Yet,, in the calculated direction feature, be inaccurate because FRAT has " around effect ".The present invention has designed a kind of improvement Radon conversion of novelty, and (Modified Finite Radon Transform abbreviates MFRAT as and calculates palm print characteristics accurately.The calculating palm print characteristics of MFRAT is as follows:

Definition Z _p=0,1 ..., and p-1}, wherein p is a positive integer, for limited two-dimensional grid Z _p ²On real-valued graph of equation as f[x, y], MFRAT is defined as:

$r\left[L_k\right]={\mathrm{MFRAT}}_f\left(k\right)=\underset{(i,j){\∈}_k}{\mathrm{\Σ}}f[i,j]---\left(1\right)$

Wherein, L _kFor at two-dimensional grid Z ² _pIn, f[x, y] the straight line formed of some points:

L _k＝{(i，j)：j＝k(i-i ₀)+j ₀，i∈Z _p} (2)

In the following formula, L _kBe straight-line equation, (i ₀, j ₀) be Z ₂ ^pCentral point.K is expressed as L _kSlope.L so _kJust be expressed as through Z ^{2 p}Central point (i ₀, j ₀) straight line of Different Slope (direction).L _kAlso has another method for expressing L (θ _k), wherein, θ _KIt is angle value corresponding to k.

R (L in the formula (1) _k) expression is to the L of different directions _kCarry out integration (summation), i.e. r (L _k) represented the L of different directions _kEnergy.By comparing r (L _k) size calculate the directional information of palmmprint.Because in the palmprint image, the pixel value of palm line is generally less, selects r (L so _k) in the direction of minimum value as final directional information.Formula as follows:

${\mathrm{\θ}}_{k(i_0,j_0)}=\arg \left({\min }_k\left(r\right[L_k\left]\right)\right),k=\mathrm{1,2},\·\·\·N---\left(3\right)$

In whole palmprint image, by the mobile Z of pixel or a plurality of pixels ² _p, the directional information of entire image is just calculated so.The directional diagram formula of palmprint image is:

Wherein (i j) is (3) formula θ to k _k(i, k value j).

In MFRAT, there are three parameters in application, to be adjusted, be respectively p, N and W.The one, p, it has determined two-dimensional grid Z ² _pSize, also just equal to have determined L _kLength; The 2nd, the quantity N of k, the energy of many fewer striplines is calculated in its expression, if N is greatly then calculated amount is big, direction character is very few if N is little, generally speaking, the size of N is between 6～12; The 3rd, L _kWidth W, can adjust the size of W according to application demand, generally speaking, the size of W is between 1～4.Among the present invention, p is set to 16; N is set to 6; W is set to 4, and final palm print characteristics image is that palmmprint training ROI image is 32 * 32 pixels.

Step (5) is based on the palmmprint coupling of point to the zone

In other palm grain identification methods, normalization Hamming distance (the Normalized HammingDistance) or angular distance (Angular Distance) often are used to characteristic matching.But be to use the matching result of Hamming distance or angular distance often healthy and strong inadequately, reason is that they are based on pixel to pixel matching.Generally speaking, owing to have displacement, rotation error between palmprint image to be identified and palmmprint training image, therefore the pixel of two images can't inregister.The distance function based on putting the zone of the present invention's design carries out the palmmprint coupling, can effectively improve the precision of palmmprint coupling.

If A is a width of cloth palmmprint training image, B is a width of cloth palmmprint test pattern, and they are still gathered in different time sections from same palm.The size of A and B all is m * n pixel.Further establish A (i, j) with B (x, y) be two in the same position corresponding point.If do not have displacement, rotation error between A and B, we know that (i, j) (x y) overlaps A, i.e. " i=x " and " j=y " with B so.But, mention as leading portion because displacement and rotation error, A (i, j) often and B (x y) does not overlap.On the other hand, (i, (x, y) near likelihood ratio is bigger j) to appear at B for A.According to above analysis, the coupling based on putting the zone of design can be expressed as:

$s(A,B)=(\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}A(i,j)\⊗\stackrel{\‾}{B}(i,j))/m\×n---\left(5\right)$

(5) in the formula, s (A, B) matching distance of expression from A to B."  " presentation logic " etc. " operation, promptly A (i, j) with B (i, j) value of any one pixel in equates, then (i, j) (i, value j) is 1 to  B to A, otherwise then is 0.(i is that (i j) is the regional area at center, can be defined as different shapes with B j) to B.Similarly, the matching distance from B to A is:

$s(B,A)=(\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}B(i,j)\⊗\stackrel{\‾}{A}(i,j))/m\×n---\left(6\right)$

Final matching distance is:

S(A，B)＝S(B，A)＝Max(s(A，B)，s(B，A)) (7)

Use the coupling of point to the zone.Among the present invention, B (i, j) be set to the cross area that area is 5 pixel sizes (B (and i-1, j), B (i+1, j), and B (i, j), B (i, j-1), B (i, j+1)), perhaps the square region of 9 pixel sizes ((B (and i-1, j-1), B (i-1, j+1), and B (i-1, j), B (i+1, j), B (i, j), B (i, j-1), B (i, j+1), and B (i+1, j+1), B (i+1, j-1)).

With respect to prior art, the invention has the beneficial effects as follows:

One uses the method for structure palmmprint training ROI image to compensate rotation error.

Mentioned as preamble background technology of the present invention, because operation is handled in incomplete pre-service, palmmprint test pattern to be identified often has bigger rotation error with the palmmprint training image of registration phase collection, the identification that easily makes the mistake, the present invention are constructed new palmmprint training ROI image and are used to compensate rotation error.For example, have width of cloth palmmprint training ROI image A in the system, palmmprint training ROI image A is rotated, angle [alpha] is respectively 3 °, 6 °, 9 ° ,-3 ° ,-6 ° ,-9 °, obtains new palmmprint training ROI image set A ₁, A ₂, A ₃, A ₄, A ₅, A ₆If there is several palmmprint training ROI image in the system, except palmmprint training ROI image A, also have B, C etc., so last palmmprint training ROI image set is A, A ₁, A ₂, A ₃, A ₄, A ₅, A ₆, B, C.Anglec of rotation α wherein, the quantity of the new palmmprint training ROI image of generation is according to the actual conditions adjustment.Method by structure palmmprint training ROI image set compensates rotation error, can effectively reduce the rotation error negative effect that identification brings to palmprint image.

They are two years old, palmmprint direction character coding RPOC is proposed, be Robust Palmprint OrientationCode, wherein improve limited Radon conversion MFRAT (Modified Finite RadonTransform), can extract the directional characteristic high precision discrimination of palmmprint fast and accurately.In the prior art, the classical palm grain identification method Competitive Code based on directional information uses the Gabor wave filter of 6 directions that image is carried out filtering, determines the direction value of pixel by the size that compares the filter response value.But the convolution of Gabor filtering and image is more time-consuming, because need use a large amount of multiplication and add operation, and the Gabor wave filter can not well be simulated line feature.

The present invention proposes improved limited Radon conversion MFRAT, in the subrange of palmprint image, except high-precision discrimination, the big advantage of palmmprint direction character coding RPOC is to have very fast processing speed, because when using MFRAT to do feature extraction, main use additive operation, add operation is carried out in wire zone in 6 directions, can effectively reduce the time that palm print characteristics extracts, and the extraordinary line feature that fits of energy has than Gabor wave filter and better extracts directional characteristic ability.In Fig. 5 of embodiment MFRAT, the width W of line is 1, and the quantity of direction is 6, (a) (b) (c) (d) (e) (f) image represented the integration (summation) of the line of different directions, these directions are respectively 0 °, and 30 °, 60 °, 90 °, 120 °, 150 °.

Be in the comparison diagram of the present invention at embodiment Fig. 7 in addition, (a) image is original palmmprint ROI image, (b) be based on the palm grain identification method of CompetitiveCode, (c) be palmmprint direction character coding RPOC, as can be seen, palmmprint direction character coding RPOC of the present invention can better reflect the palmmprint architectural feature.In Figure 11, showed the ROC curve map of the palm grain identification method (Palmcode, Fuioncode, Competitivecode) of palmmprint direction character coding RPOC and several classics.In the ROC curve, a given FAR value, the value of GAR is big more, illustrates that discrimination is good more.As can be seen, the recognition performance of palmmprint direction character coding RPOC will be significantly better than the palm grain identification method of other several classics in the prior art.

Its three, the point that proposes a kind of novelty has better fault-tolerant ability to the palm print characteristics coupling in zone.In the prior art, normalization Hamming distance (Normalized Hamming distance) or angular distance (Angular distance) often are used to characteristic matching.But be to use the matching result of Hamming distance or angular distance often healthy and strong inadequately, reason is that they are based on pixel to pixel matching.Generally speaking, owing to have displacement, rotation error between the palmmprint training image of palmmprint test pattern to be identified and registration phase collection, the pixel of the palmmprint training image that palmmprint test pattern therefore to be identified and registration phase are gathered can't inregister.

That designs among the present invention can be expressed as based on the coupling of point to the zone:

$s(A,B)=(\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}A(i,j)\⊗\stackrel{\‾}{B}(i,j))/m\×n---\left(8\right)$

(8) in the formula, s (A, B) matching distance of expression from A to B."  " presentation logic " etc. " operation, promptly A (i, j) with B (i, j) value of any one pixel in equates, then (i, j) (i, value j) is 1 to  B to A, otherwise then is 0.(i is that (i j) is the regional area at center, can be defined as different shapes with B j) to B.Similarly, the matching distance from B to A is:

$s(B,A)=(\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}B(i,j)\⊗\stackrel{\‾}{A}(i,j))/m\×n---\left(9\right)$

Final matching distance is:

S(A，B)＝S(B，A)＝Max(s(A，B)，s(B，A)) (10)

As a further improvement of existing technologies, the B among the present invention (i, j) be set to the cross area that area is 5 pixel sizes (B (and i-1, j), B (i+1, j), B (i, j), B (i, j-1), B (i, j+1), B (i, j), B (i, j-1), B (i, j+1)), perhaps the square region of 9 pixel sizes ((B (and i-1, j-1), B (i-1, j+1), B (i-1, j), B (i+1, j), B (i, j), B (i, j-1), B (i, j+1), B (i+1, j+1), B (i+1, j-1).Therefore the distance function based on putting the zone of the present invention's design carries out the palmmprint coupling, has very strong fault-tolerant ability, can effectively improve the precision of coupling.

Description of drawings

Fig. 1 is based on the process flow diagram of directional characteristic palm grain identification method.

Fig. 2 is the palmprint image that the present invention gathers.

Fig. 3 is that the present invention shears the original palmmprint training ROI image in back.

Fig. 4 is the synoptic diagram that the present invention constructs palmmprint training ROI image set.

Fig. 5 is the MFRAT synoptic diagram of the present invention's 9 * 9 sizes.

Fig. 6 is the MFRAT synoptic diagram of the present invention's 16 * 16 sizes.

Fig. 7 is RPOC of the present invention and Competitivecode feature comparison diagram.

Fig. 8 is the synoptic diagram of point of the present invention to zone coupling operator.

Fig. 9 is in the demonstration test of the present invention, true coupling and the false matching value distribution plan that mates.

Figure 10 is the FAR and the FRR distribution plan of demonstration test of the present invention.

Figure 11 is a palmmprint direction character coding RPOC result of the present invention ROC curve map relatively.

Embodiment is further described embodiments of the invention below in conjunction with accompanying drawing.

Fig. 1 is based on the process flow diagram of directional characteristic palm grain identification method.In Fig. 1, comprise registration process and identifying based on directional characteristic palm grain identification method.

Registration process is: the user uses collecting device to carry out palm-print image capture, and deposits in the system user's personal information such as name, ID number etc.Generally speaking, need carry out 1～3 time to every user and gather, deposit masterplate database 1～3 width of cloth palmmprint training image in.Adopt the palmprint image pre-service, the central area of 128 * 128 sizes is as palmmprint training ROI image in the cutting palmmprint training image.Structure palmmprint training ROI image set is to train the ROI image to carry out some low-angles rotations to a secondary palmmprint wherein, forms some postrotational palmmprints training ROI images, promptly forms new palmmprint training ROI image set.Use MFRAT to extract the direction character of all palmmprint training ROI images, the palmmprint direction character masterplate that forms palmmprint training plan image set is deposited in the masterplate database.Wherein, the palmmprint direction character masterplate of each palmmprint training ROI image set all and correspondence such as the personally identifiable information that had before deposited in such as name, ID number.

Identifying is: the user uses collecting device to carry out palm-print image capture.To the pre-service of palmmprint test pattern, the central area of 128 * 128 sizes is as palmmprint test ROI image in the cutting palmmprint test pattern.Extract the direction character that palmmprint is tested the ROI image with MFRAT, form palmmprint direction character masterplate to be identified.For authentication operation, user to be identified also needs to import ID number and waits identity information in system.Identical ID number the some palmmprints training direction character masterplates of having in palmmprint measurement direction feature masterplate and the training masterplate database mate.Use the palmmprint coupling of point to the zone.Get and train the direction character masterplate to have the value of maximum similarity in measurement direction feature masterplate and the masterplate database as final matching value.If this matching value is greater than prior preset threshold T, then verification operation is successful.Otherwise authentication failed.Operate for identity identification, user to be identified need not to import ID number and waits identity information in system, all training characteristics masterplates mate in palmmprint measurement direction feature masterplate and the training masterplate database, system returns ID number of training direction character masterplate that has the maximum match value with measurement direction feature masterplate, if this matching value is greater than prior preset threshold T, then identification is successful.Otherwise identification failure.

Fig. 2 is the palmprint image that the present invention gathers.In the palmprint image of gathering, background is a black, is convenient to be partitioned into the palmmprint area image.This image resolution ratio is about 75dpi.Though resolution is low, features such as palmmprint main line, gauffer are still quite clear, can be used for authentication.Using first advantage of low resolution palmprint image is that image capture device is cheap, is beneficial to reduce cost, and second advantage is that size of images is little, and speed is fast when handling.

Fig. 3 is that the present invention shears the original palmmprint training ROI image in back.In the palmmprint training ROI image after shearing, still contain the principal character of palmmprint, as main line, gauffer etc.Using the fundamental purpose of palmmprint training ROI image is that palmprint image is positioned, and makes to have smaller displacement and rotation error from same palm between the palmprint image of different time collection, is convenient to the palmmprint coupling.

Fig. 4 is the synoptic diagram that the present invention constructs palmmprint training ROI image set.By observing, palmprint image to be identified is that the maximum rotation error between the palmmprint training image of test pattern and registration phase collection is about 10 °.If to someone's palmmprint, there is width of cloth palmmprint training ROI image A in the system, palmmprint training ROI image A is rotated 3 °, 6 °, 9 ° ,-3 ° ,-6 ° ,-9 ° respectively, obtain new training palmmprint ROI image A ₁, A ₂, A ₃, A ₄, A ₅, A ₆If there is several palmmprint training ROI image in the system, except palmmprint training ROI image A, also have B, C etc., so last palmmprint training plan image set is A, A ₁, A ₂, A ₃, A ₄, A ₅, A ₆, B, C.Structure by palmmprint training ROI image set can effectively compensate rotation error.

Fig. 5 is the MFRAT synoptic diagram of the present invention's 9 * 9 sizes.In Fig. 5 MFRAT, the width W of line is 1, and the quantity of direction is 6, (a) (b) (c) (d) (e) (f) image represented the integration (summation) of the line of different directions, these directions are respectively 0 °, 30 °, 60 °, 90 °, 120 °, 150 °.Use this MFRAT, can calculate the direction character of a pixel (center pixel) at every turn.

Fig. 6 is the MFRAT synoptic diagram of the present invention's 16 * 16 sizes.In the MFRAT of Fig. 6, the width W of line is 4, and the quantity of direction is 6, (a) (b) (c) (d) (e) (f) image represented the integration of the line of different directions, these directions are respectively 0 °, 30 °, 60 °, 90 °, 120 °, 150 °.Use this MFRAT, can calculate the direction character of 4 * 4 pixels (dark pixels of central area) at every turn.These 4 * 4 pixels have identical direction character value, and in characteristic image, these 4 * 4 pixels can be looked at as 1 pixel, so 128 * 128 original image, and its characteristic image is 32 * 32 pixel sizes.

Fig. 7 is RPOC of the present invention and Competitivecode feature comparison diagram.Wherein (a) image is original palmmprint ROI image, (b) is based on the palm grain identification method of CompetitiveCode, (c) is palmmprint direction character coding RPOC.As can be seen, palmmprint direction character coding RPOC can better reflect the palmmprint architectural feature.Here it is pointed out that Fig. 7 (b) and size (c) are 32 * 32 pixels, show that for convenience they are exaggerated.Wherein each lattice is represented a pixel.In addition, Fig. 7 (b) with (c) in, different gray-scale values has been represented different direction values.

Fig. 8 is the synoptic diagram of point of the present invention to zone coupling operator.Wherein (a) is point-to-point coupling operator, (b) is the coupling operator of point to cross area.(c) be that point is to little square coupling operator.

Fig. 9 is in the demonstration test of the present invention, the distribution plan of matching value.

In confirmatory experiment, be called true coupling from the palmmprint of same palm coupling, i.e. Genuine coupling is called false coupling from the palmmprint coupling of different palms, i.e. the Imposter coupling.The result of true coupling is exactly the checking of validated user to own legal identity.False coupling then is the coupling of pretending to be real user.Fig. 9 is the statistics of matching value during all true couplings are mated with vacation, and matching value distribution center's point of true coupling is about 0.7, and the false matching value distribution center's point that mates is about 0.5.As can be seen from Figure 9, true coupling can be separated preferably with false matching value, and the jactitator is difficult to pretend to be successfully.For a perfect system, the distribution of true matching value and false matching value should not have intersection point.

Figure 10 is the FAR and the FRR distribution plan of demonstration test of the present invention.

Generally in the biometrics identification technology adopt three indexs to weigh recognition effect, promptly misclassification rate (False Acceptance Rate, FAR), mistake refuse rate (False Reject Rate, FRR) and etc. wrong rate (Equal Error Rate, EER).FRR is meant system is refused real user as the personator probability; FAR is meant system accepts the personator as real user probability.EER is meant the error rate when FAR equates with FRR.FRR and FAR are two parameters of same algorithmic system, and they are placed in the same coordinate, and as shown in the figure, FAR increases along with threshold value and reduces, and FRR increases and increases along with threshold value.Generally speaking, FAR and FRR are inverse relation, and FAR is big more, and then FRR is more little, and vice versa.FAR and FRR have the point of crossing in the drawings, and this point is the point (equivalent point of corresponding this threshold value is called EER) of FAR and FRR equivalence under certain threshold value.The combination property of using EER to come measure algorithm traditionally for a more excellent palm-print identifying arithmetic, wishes that FAR and FRR are the smaller the better under the same threshold situation.

Figure 11 is a palmmprint direction character coding RPOC result of the present invention ROC curve map relatively.In Figure 11, showed the ROC curve map of the palm grain identification method (Palmcode, Fuioncode, Competitivecode) of palmmprint direction character coding RPOC and several classics.In the ROC curve, horizontal ordinate is misclassification rate (FAR), ordinate be correct receptance (Genuine Acceptance Rate, GAR).In the ROC curve, a given misclassification rate FAR value, the value of correct receptance GAR is big more, illustrates that discrimination is good more.From Figure 11, as can be seen, the recognition performance of palmmprint direction character coding RPOC will be significantly better than the palm grain identification method of other several classics.

Embodiment

(1) image data base

Adopt algorithm of the present invention in the palm print database of The Hong Kong Polytechnic University living things feature recognition research centre (PolyU_BRC), to test.These images are at twice the masculinity and femininity of all ages and classes to be gathered, and twice acquisition interval average out to 2 months gathered palmprint image about 10 width of cloth to a palm at every turn.So each palm has nearly 20 width of cloth palmprint images in the database.The palmprint image size is 384 * 284 pixels.

(2) image pre-service

Palmmprint Preprocessing Algorithm [the list of references: D.Zhang of people's propositions such as roc is magnified in application, W.K.Kong, J.You, and M.Wong, " Online palmprint identification, " IEEETransactions on Pattern Analysis and Machine Intelligence, 25 (9) (2003), pp.1041-1050.], intercepting palmprint image center size is the palmmprint ROI image block of 128 * 128 pixels, carries out palm print characteristics and extract and coupling on this palmmprint ROI image block.

(3) structure palmmprint training ROI image set

We select for use first width of cloth image of each palm of collection to train the ROI image set as palmmprint.19 remaining width of cloth palmprint images are as palmmprint test pattern image set.If, palmmprint training ROI image A is rotated 3 °, 6 °, 9 ° ,-3 ° ,-6 ° ,-9 ° respectively, obtain new palmmprint training ROI image set A to someone's palmmprint training ROI image A ₁, A ₂, A ₃, A ₄, A ₅, A ₆So last palmmprint training ROI image set is A, A ₁, A ₂, A ₃, A ₄, A ₅, A ₆, have 7 width of cloth palmmprints training ROI image.

(4) use MFRAT to extract the direction character of palmmprint

Use MFRAT to extract the direction character of palmmprint, among the present invention, p is set to 16; N is set to 6; W is set to 4.Final characteristic image is that palmmprint training ROI image size is 32 * 32 pixels.

(5) use is mated based on the palmmprint of point to the zone

Use is based on the palmmprint coupling of point to the zone.

Among the present invention, B (i, j) be set to the cross area that area is 5 pixels (B (and i-1, j), B (i+1, j), and B (i, j), B (i, j-1), B (i, j+1)), perhaps the square region of 9 pixels ((B (and i-1, j-1), B (i-1, j+1), and B (i-1, j), B (i+1, j), B (i, j), B (i, j-1), B (i, j+1), and B (i+1, j+1), B (i+1, j-1)).

(6) palmmprint identification test interpretation of result

The palmmprint identification test can be divided into two classes, i.e. checking (Verification) and identification (Identification).Checking is carried out man-to-man comparison (one-to-one matching) by palmmprint and palmmprint of having registered that collects exactly, confirms the process of identity.As the precondition of checking, his or her palmmprint must be registered in the palmmprint storehouse.Palmmprint is with the storage of certain compressed format, and (ID PIN) connects with its name or its sign.On-the-spot in comparison subsequently, its sign of checking earlier then, is utilized the palmmprint of the palmmprint of system and collection in worksite to compare and is proved that its sign is legal.Checking is to have answered such problem in fact: " he is this people that he calls oneself? "

Identification then is that the palmmprint that collects is contrasted one by one with the palmmprint in the palm print database, therefrom finds out and test the fingerprint that palmmprint is complementary.This also is " one-to-many coupling (one-to-manymatching) ".Identification is to have answered such problem in fact: " who is he? "

1, checking interpretation of result

In demonstration test, the palmmprint test pattern concentrates all palmprint images to concentrate all images to mate with the palmmprint training image.If palmmprint test pattern and palmmprint training image come from same palm, coupling between them is called as true coupling (Genuine Matching) so, if test pattern and palmmprint training image are from different palms, the coupling between them is called as the coupling (Impostor Matching) of assuming another's name so.The result that coupling produces be a matching value, and the scope of matching value if matching value has surpassed given threshold value, is then thought and verified and pass through, otherwise be rejected between [0,1].Fig. 9 has showed true coupling Genuine Matching and the matching value distribution plan of assuming another's name to mate ImpostorMatching.

Figure 10 and following table have been showed the inventive method FAR and FRR value when different threshold value.As can be seen, when FAR 4.0 * 10 ^-5During %, FRR only is 1.631%.EER is about 0.16% when threshold value is 0.616.

Threshold value	FAR(％)	FRR(％)
			0.577 0.578 0.590 0.600 0.606 0.612 0.616	8.530 7.951 2.637 0.979 0.502 0.245 0.148	0 0.030 0.046 0.076 0.091 0.137 0.167

?0.620 ?0.630 ?0.640 ?0.650 ?0.660 ?0.661

0.088 0.017 3.6×10 -3 4.4×10 -4 4.0×10 -5 0

?0.213 ?0.320 ?0.442 ?0.869 ?1.631 ?1.745

Figure 11 has showed that with following table the recognition result of different palm grain identification methods compares.As FAR during at 4.0 * 10-5%, the FRR of Palmcode is 17.2%, and the FRR of Fusioncode is 12.1%, and the FRR of Competitivecode is 4.86%, and the FRR of palmmprint direction character coding RPOC only is 1.631%.The EER of Palmcode is 0.98%, and the EER of Fusioncode is 0.87%, and the EER of Competitivecode is 0.47%, and the EER of RPOC only is 0.16%.Result than other several methods is well a lot.At present, from the data that can find, palmmprint direction character coding RPOC of the present invention has obtained the higher discrimination in palmmprint identification field.

	Palmcode	?Fusioncode	Competitive?code	RPOC
					FAR(％) FRR(％) EER(％)	4×10 -517.2 0.98	?4×10 -5?12.1 ?0.82	4×10 -5 4.86 0.47	4×10 -5 1.631 0.16

2, identification result analysis

As the palmmprint training image, 19 remaining width of cloth images carry out the identification test as the palmmprint test pattern with first width of cloth image of each palm.The identification precision of palmmprint direction character coding RPOC is 98.12%, and PalmCode, FusionCode, the identification precision of CompetitiveCode is respectively 95.41%, 96.46% and 97.85%.In palmmprint identification test, palmmprint direction character coding RPOC has also obtained discrimination preferably.

3, memory space

The characteristic image of palmmprint direction character coding RPOC is that palmmprint training image size is 32 * 32, and the value of each pixel may be in 1,2,3,4,5,6 these six numbers.Use 3 bits just can represent this several values to each pixel so, as use 001 to represent 1,010 to represent 2,011 to represent 3,100 to represent 4,101 to represent 5,110 to represent 6.In this way store the direction character of a width of cloth palmmprint training image, the byte number that then needs to use is (32 * 32 * 3)/8=384bytes.As seen, the memory space of palmmprint direction character coding RPOC is very little, is fit to very much use in real time.

4, processing speed

Except high-precision discrimination, another big advantage of palmmprint direction character coding RPOC is to have very fast processing speed, because when using MFRAT to do feature extraction, mainly use additive operation, has therefore reduced the time overhead of processor.All tests are to finish on dominant frequency is the PC of Pentium processor, 256 MB of memory of 2.4GHZ, and the programming platform that uses is visual c++.Following table has been listed the pre-service of palmmprint direction character coding RPOC algorithm, feature extraction and has been mated needed averaging time.Use palmmprint direction character coding RPOC method to carry out the average response time of an authentication less than 0.4 second.The feature extraction time of palmmprint direction character coding RPOC only is 50 milliseconds, and the feature extraction time of CompetitiveCode is 200 milliseconds, and the palmmprint direction character coding used feature extraction time of RPOC is 1/4th of CompetitiveCode.

	Method	Processing time
			Pretreatment time feature extraction match time time	? RPOC Competitive?Code RLOC	316ms 50ms 200ms 2.5ms

Claims (4)

1, a kind of based on directional characteristic palm grain identification method, comprise a: palm-print image capture, the user carries out palm-print image capture by harvester, obtain can be used for the further palmprint image gray matrix of processing, palmprint image in the registration phase collection is called the palmmprint training image, is called the palmmprint test pattern at the palmprint image of cognitive phase collection;

B: palmprint image pre-service, at first by location palm, finger position, palmprint image is rotated correction, train ROI (Region of Interest) image in the square region of the centre of palmprint image cutting 128 * 128 pixels as palmmprint then, the palmmprint training ROI image in last square shaped zone carries out feature extraction and coupling, it is characterized in that this method also comprises:

C: structure palmmprint training ROI image set, if palmmprint to the someone, there is width of cloth palmmprint training ROI image A in the system, to palmmprint training ROI image A respectively anglec of rotation α be 3 °, 6 °, 9 ° ,-3 ° ,-6 ° ,-9 °, obtain postrotational palmmprint training ROI image, promptly form new palmmprint training ROI image set A ₁, A ₂, A ₃, A ₄, A ₅, A ₆, also including several palmmprint training of B, C ROI image in the system of setting up departments, then last palmmprint training ROI image set is A, A ₁, A ₂, A ₃, A ₄, A ₅, A ₆, B, C, the structure by palmmprint training ROI image set can effectively compensate rotation error;

D: palm print characteristics extracts-sets up palmmprint direction character coding RPOC (RobustPalmprint Orientation Code)

The principal character of palmmprint is the line feature, and these lines have directivity, and promptly direction character can be expressed the essential structure of palmmprint, extracts directional characteristic MFRAT and is described below:

Definition Z _p=0,1 ..., and p-1}, wherein p is a positive integer, for limited two-dimensional grid Z ² _pOn real-valued Equation f [x, y], MFRAT is defined as:

$r\left[L_k\right]={\mathrm{MFRAT}}_f\left(k\right)=\underset{(i,j)\∈L_k}{\mathrm{\Σ}}f[i,j]$

Wherein, f[x, y] be the gradation of image matrix, L _kFor at two-dimensional grid Z ² _pIn, f[x, y] the straight line formed of some points:

L _k＝{(i，j)：j＝k(i-i ₀)+j ₀，i∈Z _p}

In the following formula, L _kBe straight-line equation, (i ₀, j ₀) be Z ² _pCentral point, k is expressed as L _kSlope; L so _kJust be expressed as through Z ² _pCentral point (i ₀, j ₀) straight line of different directions, L _kAlso has another method for expressing L (θ _k), wherein, θ _kIt is angle value corresponding to k;

R (L in the formula _k) expression is to the L of different directions _kCarry out integration and promptly sue for peace, r (L _k) represented the L of different directions _kEnergy; By comparing r (L _k) calculate the directional information of palmmprint; Select r (L _k) the little direction of intermediate value is as (i ₀, j ₀) final directional information; Formula as follows:

${\mathrm{\θ}}_{k(i_0,j_0)}=\arg \left({\min }_k\left(r\right[L_k\left]\right)\right),k=\mathrm{1,2},\·\·\·N$

In whole palmprint image, by the mobile Z of pixel or a plurality of pixels ² _p, the directional information of whole palmprint image is just calculated, and the directional diagram formula of palmprint image is:

Wherein (i j) is formula θ to k _{K (i, j)}The k value;

In MFRAT, there are three parameters in application, to adjust, be respectively p, N and W, wherein: p has determined two-dimensional grid Z ² _pSize, promptly determined L _kLength; The quantity N of k represents to calculate the quantity of line, if N greatly then calculated amount is big, direction character is few if N is little, and N is between 6～12; L _kWidth W can adjust according to application demand, W is between 1～4;

Use MFRAT to extract the direction character of all palmmprint training ROI images, the palmmprint direction character masterplate that forms palmmprint training ROI image set is deposited in the masterplate database;

E: based on the palmmprint coupling of point to the zone

If the A in the different time sections collection from same palm is a width of cloth palmmprint training image, B is a width of cloth palmmprint test pattern, the size of A and B all is m * n pixel, and further establishing does not have displacement and rotation error, A (i between A and B, j) with B (x, y) be two corresponding point in same position, (i is j) with B (x for A at this moment, y overlaps, i.e. " i=x " and " j=y ", since displacement and rotation error, A (i, j) often and B (x, y) do not overlap, but A (i j) appears at B (x, y) near probability is big, and the palmmprint matching list based on putting the zone of design is shown:

$s(A,B)=(\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}A(i,j)\⊗\stackrel{\‾}{B}(i,j))/m\×n$

In the formula, s (A, B) matching distance of expression from A to B, the operation of "  " presentation logic, (i is j) with B (i for A, the value of any one pixel j) is equal, then A (i, j)  B (i, j) value is 1, otherwise then be 0, (i is with B (i j) to B, j) be the regional area at center, can be defined as different shapes;

Similarly, the matching distance from B to A is:

$s(B,A)=(\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}B(i,j)\⊗\stackrel{\‾}{A}(i,j))/m\×n$

Final matching distance is:

S(A，B)＝S(B，A)＝Max(s(A，B)，s(B，A))。

2, according to claim 1 based on directional characteristic palm grain identification method, it is characterized in that: the described palmmprint of establishing the someone, there is width of cloth palmmprint training ROI image A in the system, to palmmprint training ROI image A respectively anglec of rotation α be 3 °, 6 °, 9 ° ,-3 ° ,-6 ° ,-9 °, obtain postrotational palmmprint training ROI image, promptly form new palmmprint training ROI image set A ₁, A ₂, A ₃, A ₄, A ₅, A ₆, also including several palmmprint training of B, C ROI image in the system of setting up departments, then last palmmprint training ROI image set is A, A ₁, A ₂, A ₃, A ₄, A ₅, A ₆, B, C, anglec of rotation α wherein, the quantity that generates new palmmprint training ROI image is according to the actual conditions adjustment.

3, according to claim 1 based on directional characteristic palm grain identification method, it is characterized in that: described in MFRAT, there are three parameters in application, to adjust, be respectively p, N and W, wherein p is set at 16, N is set at 6, W is set at 4, and final palm print characteristics image is that palmmprint training ROI image is 32 * 32 pixels.

4, according to claim 1 based on directional characteristic palm grain identification method, it is characterized in that: described based on the palmmprint coupling of point to the zone, wherein (i j) is set at the cross area that area is 5 pixels (B (i-1 to B, j), and B (i+1, j), B (i, j), B (i, j-1), B (i, j+1)), perhaps the square region of 9 pixels ((B (and i-1, j-1), B (i-1, j+1), and B (i-1, j), B (i+1, j), B (i, j), and B (i, j-1), B (i, j+1), B (i+1, j+1), B (i+1, j-1)).

Images